{"sample_id": "sample_00000001", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0015_00", "gt_frame_id": "755", "anchor_frame_id": "827", "shot": 6, "contexts_frame_id": ["827", "1724", "1704", "12", "1620", "1918"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000013", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0015_00", "gt_frame_id": "1152", "anchor_frame_id": "1220", "shot": 6, "contexts_frame_id": ["1220", "242", "1778", "1368", "1392", "1422"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000019", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0025_02", "gt_frame_id": "471", "anchor_frame_id": "436", "shot": 6, "contexts_frame_id": ["436", "338", "272", "1180", "1348", "1376"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000022", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0030_00", "gt_frame_id": "1626", "anchor_frame_id": "1726", "shot": 6, "contexts_frame_id": ["1726", "2040", "1886", "1948", "1668", "2496"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000025", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0030_01", "gt_frame_id": "59", "anchor_frame_id": "203", "shot": 6, "contexts_frame_id": ["203", "644", "564", "608", "1610", "34"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000028", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0030_01", "gt_frame_id": "1389", "anchor_frame_id": "1510", "shot": 6, "contexts_frame_id": ["1510", "1482", "1548", "1610", "1416", "1332"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000031", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0030_02", "gt_frame_id": "508", "anchor_frame_id": "380", "shot": 6, "contexts_frame_id": ["380", "328", "432", "478", "538", "678"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000034", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0030_02", "gt_frame_id": "805", "anchor_frame_id": "518", "shot": 6, "contexts_frame_id": ["518", "476", "428", "694", "754", "858"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000037", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0030_02", "gt_frame_id": "520", "anchor_frame_id": "803", "shot": 6, "contexts_frame_id": ["803", "1764", "742", "682", "556", "482"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000040", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0046_00", "gt_frame_id": "1817", "anchor_frame_id": "1563", "shot": 6, "contexts_frame_id": ["1563", "1664", "1692", "1516", "2058", "2004"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000043", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0050_02", "gt_frame_id": "396", "anchor_frame_id": "347", "shot": 6, "contexts_frame_id": ["347", "2266", "4094", "3062", "3028", "1098"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000046", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0050_02", "gt_frame_id": "76", "anchor_frame_id": "107", "shot": 6, "contexts_frame_id": ["107", "1370", "1494", "3750", "3792", "3304"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000049", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0050_02", "gt_frame_id": "3115", "anchor_frame_id": "3077", "shot": 6, "contexts_frame_id": ["3077", "1948", "4254", "3908", "2574", "310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000055", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0050_02", "gt_frame_id": "122", "anchor_frame_id": "53", "shot": 6, "contexts_frame_id": ["53", "1416", "1348", "3746", "1638", "3396"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000058", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0050_02", "gt_frame_id": "281", "anchor_frame_id": "310", "shot": 6, "contexts_frame_id": ["310", "3112", "3084", "2494", "3828", "1630"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000061", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0050_02", "gt_frame_id": "3140", "anchor_frame_id": "3250", "shot": 6, "contexts_frame_id": ["3250", "256", "292", "2866", "3786", "3728"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000070", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0088_01", "gt_frame_id": "634", "anchor_frame_id": "698", "shot": 6, "contexts_frame_id": ["698", "1056", "674", "372", "514", "584"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000073", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0088_02", "gt_frame_id": "1675", "anchor_frame_id": "1640", "shot": 6, "contexts_frame_id": ["1640", "2198", "2162", "1088", "932", "1712"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000076", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0088_03", "gt_frame_id": "163", "anchor_frame_id": "137", "shot": 6, "contexts_frame_id": ["137", "670", "1186", "1444", "630", "54"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000079", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0088_03", "gt_frame_id": "139", "anchor_frame_id": "167", "shot": 6, "contexts_frame_id": ["167", "1182", "1204", "1238", "1284", "96"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000088", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0095_01", "gt_frame_id": "651", "anchor_frame_id": "689", "shot": 6, "contexts_frame_id": ["689", "748", "922", "598", "988", "428"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000094", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0131_01", "gt_frame_id": "607", "anchor_frame_id": "546", "shot": 6, "contexts_frame_id": ["546", "566", "650", "678", "728", "822"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000097", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0144_00", "gt_frame_id": "204", "anchor_frame_id": "121", "shot": 6, "contexts_frame_id": ["121", "660", "164", "1290", "570", "248"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000100", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0144_00", "gt_frame_id": "120", "anchor_frame_id": "204", "shot": 6, "contexts_frame_id": ["204", "160", "1306", "644", "880", "692"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000103", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0146_02", "gt_frame_id": "169", "anchor_frame_id": "376", "shot": 6, "contexts_frame_id": ["376", "512", "582", "318", "292", "96"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000106", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0149_00", "gt_frame_id": "1211", "anchor_frame_id": "1173", "shot": 6, "contexts_frame_id": ["1173", "306", "680", "1366", "1336", "1248"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000109", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0164_00", "gt_frame_id": "1380", "anchor_frame_id": "1469", "shot": 6, "contexts_frame_id": ["1469", "1486", "228", "1558", "542", "670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000112", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_00", "gt_frame_id": "1600", "anchor_frame_id": "1469", "shot": 6, "contexts_frame_id": ["1469", "1484", "234", "1566", "576", "778"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000121", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_00", "gt_frame_id": "200", "anchor_frame_id": "76", "shot": 6, "contexts_frame_id": ["76", "1184", "1848", "1282", "238", "274"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000124", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_00", "gt_frame_id": "1469", "anchor_frame_id": "1600", "shot": 6, "contexts_frame_id": ["1600", "488", "1540", "324", "304", "1410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000127", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0164_00", "gt_frame_id": "1244", "anchor_frame_id": "1392", "shot": 6, "contexts_frame_id": ["1392", "1432", "224", "1274", "1202", "1164"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000130", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_01", "gt_frame_id": "468", "anchor_frame_id": "541", "shot": 6, "contexts_frame_id": ["541", "586", "390", "336", "298", "256"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000133", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0164_01", "gt_frame_id": "503", "anchor_frame_id": "591", "shot": 6, "contexts_frame_id": ["591", "550", "386", "418", "336", "298"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000136", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_01", "gt_frame_id": "488", "anchor_frame_id": "456", "shot": 6, "contexts_frame_id": ["456", "424", "522", "532", "272", "240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000139", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_02", "gt_frame_id": "113", "anchor_frame_id": "214", "shot": 6, "contexts_frame_id": ["214", "1142", "1082", "280", "364", "484"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000142", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0164_03", "gt_frame_id": "128", "anchor_frame_id": "349", "shot": 6, "contexts_frame_id": ["349", "266", "200", "1100", "408", "454"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000145", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0196_00", "gt_frame_id": "1212", "anchor_frame_id": "1481", "shot": 6, "contexts_frame_id": ["1481", "1246", "304", "454", "916", "1176"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000148", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0196_00", "gt_frame_id": "1481", "anchor_frame_id": "1212", "shot": 6, "contexts_frame_id": ["1212", "1224", "1344", "1376", "708", "854"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000151", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0217_00", "gt_frame_id": "412", "anchor_frame_id": "458", "shot": 6, "contexts_frame_id": ["458", "1220", "374", "492", "548", "582"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000154", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0217_00", "gt_frame_id": "621", "anchor_frame_id": "574", "shot": 6, "contexts_frame_id": ["574", "592", "646", "532", "478", "1234"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000157", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0231_00", "gt_frame_id": "2090", "anchor_frame_id": "2117", "shot": 6, "contexts_frame_id": ["2117", "2182", "2358", "2282", "2616", "3426"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000160", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_00", "gt_frame_id": "2485", "anchor_frame_id": "2255", "shot": 6, "contexts_frame_id": ["2255", "3816", "3872", "3700", "2450", "2762"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.9 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000163", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0231_00", "gt_frame_id": "2467", "anchor_frame_id": "2741", "shot": 6, "contexts_frame_id": ["2741", "2668", "3722", "2414", "2258", "3638"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000166", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_00", "gt_frame_id": "4362", "anchor_frame_id": "4420", "shot": 6, "contexts_frame_id": ["4420", "4388", "1598", "214", "1110", "612"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000169", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_00", "gt_frame_id": "2252", "anchor_frame_id": "2486", "shot": 6, "contexts_frame_id": ["2486", "2636", "2992", "3820", "3670", "2414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move backward about 1.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000175", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_00", "gt_frame_id": "3419", "anchor_frame_id": "3356", "shot": 6, "contexts_frame_id": ["3356", "4142", "1374", "1324", "1298", "3932"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000178", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_00", "gt_frame_id": "3825", "anchor_frame_id": "3884", "shot": 6, "contexts_frame_id": ["3884", "3866", "2234", "2358", "2398", "2440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000181", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0231_00", "gt_frame_id": "1440", "anchor_frame_id": "1471", "shot": 6, "contexts_frame_id": ["1471", "684", "3356", "3392", "3272", "4370"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000184", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0231_01", "gt_frame_id": "2771", "anchor_frame_id": "2559", "shot": 6, "contexts_frame_id": ["2559", "2812", "2576", "3064", "2900", "1600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000187", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0231_01", "gt_frame_id": "1437", "anchor_frame_id": "1323", "shot": 6, "contexts_frame_id": ["1323", "610", "3406", "3008", "1480", "3532"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.2 meters, then move right about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000193", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_01", "gt_frame_id": "1163", "anchor_frame_id": "1071", "shot": 6, "contexts_frame_id": ["1071", "3722", "442", "3234", "3486", "4226"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move right about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000199", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0231_01", "gt_frame_id": "1326", "anchor_frame_id": "1441", "shot": 6, "contexts_frame_id": ["1441", "1554", "4140", "628", "4210", "2926"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.1 meters, then move left about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000202", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0231_01", "gt_frame_id": "1586", "anchor_frame_id": "1622", "shot": 6, "contexts_frame_id": ["1622", "3940", "2474", "2418", "1476", "2988"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000205", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_01", "gt_frame_id": "3451", "anchor_frame_id": "3423", "shot": 6, "contexts_frame_id": ["3423", "4134", "4142", "3288", "2106", "1714"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000208", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_01", "gt_frame_id": "551", "anchor_frame_id": "592", "shot": 6, "contexts_frame_id": ["592", "624", "4226", "4264", "2054", "4308"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move left about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000211", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_01", "gt_frame_id": "2548", "anchor_frame_id": "2501", "shot": 6, "contexts_frame_id": ["2501", "2786", "3894", "3956", "1578", "4060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000217", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_02", "gt_frame_id": "1028", "anchor_frame_id": "989", "shot": 6, "contexts_frame_id": ["989", "2028", "1938", "2992", "1086", "1176"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000220", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0231_02", "gt_frame_id": "2562", "anchor_frame_id": "2666", "shot": 6, "contexts_frame_id": ["2666", "2692", "2720", "2612", "236", "206"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000226", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0246_00", "gt_frame_id": "486", "anchor_frame_id": "400", "shot": 6, "contexts_frame_id": ["400", "30", "2946", "162", "98", "2620"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000229", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0246_00", "gt_frame_id": "339", "anchor_frame_id": "282", "shot": 6, "contexts_frame_id": ["282", "30", "18", "374", "1218", "1184"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000235", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0249_00", "gt_frame_id": "310", "anchor_frame_id": "274", "shot": 6, "contexts_frame_id": ["274", "1000", "934", "340", "1262", "1306"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000244", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0249_00", "gt_frame_id": "273", "anchor_frame_id": "318", "shot": 6, "contexts_frame_id": ["318", "364", "422", "1954", "982", "214"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000247", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0249_00", "gt_frame_id": "2466", "anchor_frame_id": "2351", "shot": 6, "contexts_frame_id": ["2351", "712", "550", "510", "2652", "2568"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000250", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0251_00", "gt_frame_id": "1188", "anchor_frame_id": "1270", "shot": 6, "contexts_frame_id": ["1270", "1386", "1224", "1110", "818", "760"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000253", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0256_02", "gt_frame_id": "971", "anchor_frame_id": "1081", "shot": 6, "contexts_frame_id": ["1081", "940", "1016", "872", "666", "636"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move forward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000256", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0256_02", "gt_frame_id": "1108", "anchor_frame_id": "946", "shot": 6, "contexts_frame_id": ["946", "906", "74", "574", "532", "448"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.1 meters, then move right about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000259", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0277_01", "gt_frame_id": "174", "anchor_frame_id": "277", "shot": 6, "contexts_frame_id": ["277", "218", "812", "748", "392", "424"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000262", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0277_01", "gt_frame_id": "275", "anchor_frame_id": "190", "shot": 6, "contexts_frame_id": ["190", "214", "350", "370", "418", "456"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000268", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0300_00", "gt_frame_id": "1147", "anchor_frame_id": "904", "shot": 6, "contexts_frame_id": ["904", "802", "1254", "1642", "1678", "166"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000271", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0300_00", "gt_frame_id": "1654", "anchor_frame_id": "1586", "shot": 6, "contexts_frame_id": ["1586", "148", "214", "284", "34", "1006"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000274", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0300_00", "gt_frame_id": "901", "anchor_frame_id": "1190", "shot": 6, "contexts_frame_id": ["1190", "1022", "858", "1672", "1564", "1534"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000277", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0300_01", "gt_frame_id": "74", "anchor_frame_id": "162", "shot": 6, "contexts_frame_id": ["162", "264", "1698", "998", "1064", "108"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000280", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0300_01", "gt_frame_id": "1409", "anchor_frame_id": "1596", "shot": 6, "contexts_frame_id": ["1596", "1574", "214", "1638", "1704", "852"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000283", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0300_01", "gt_frame_id": "1121", "anchor_frame_id": "970", "shot": 6, "contexts_frame_id": ["970", "558", "1048", "58", "434", "744"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000289", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0300_01", "gt_frame_id": "1001", "anchor_frame_id": "1113", "shot": 6, "contexts_frame_id": ["1113", "1156", "1044", "28", "530", "612"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000292", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0304_00", "gt_frame_id": "288", "anchor_frame_id": "177", "shot": 6, "contexts_frame_id": ["177", "136", "1200", "1090", "318", "606"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000295", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0304_00", "gt_frame_id": "145", "anchor_frame_id": "299", "shot": 6, "contexts_frame_id": ["299", "278", "1200", "1276", "114", "732"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000298", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0307_00", "gt_frame_id": "135", "anchor_frame_id": "175", "shot": 6, "contexts_frame_id": ["175", "1970", "762", "0", "212", "230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000301", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0307_00", "gt_frame_id": "126", "anchor_frame_id": "215", "shot": 6, "contexts_frame_id": ["215", "1996", "1968", "1954", "24", "594"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000304", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0307_00", "gt_frame_id": "1110", "anchor_frame_id": "1246", "shot": 6, "contexts_frame_id": ["1246", "1290", "1212", "1168", "1142", "1042"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000307", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0307_01", "gt_frame_id": "1955", "anchor_frame_id": "1803", "shot": 6, "contexts_frame_id": ["1803", "108", "3130", "3160", "642", "804"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000313", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0307_01", "gt_frame_id": "3115", "anchor_frame_id": "3148", "shot": 6, "contexts_frame_id": ["3148", "2372", "2398", "1820", "1794", "3060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000316", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0307_01", "gt_frame_id": "520", "anchor_frame_id": "447", "shot": 6, "contexts_frame_id": ["447", "1706", "118", "3066", "3122", "1824"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000319", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0307_02", "gt_frame_id": "1409", "anchor_frame_id": "1440", "shot": 6, "contexts_frame_id": ["1440", "1434", "1384", "1376", "912", "1028"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000325", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0307_02", "gt_frame_id": "179", "anchor_frame_id": "107", "shot": 6, "contexts_frame_id": ["107", "42", "468", "134", "580", "2068"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000328", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0328_00", "gt_frame_id": "1255", "anchor_frame_id": "1229", "shot": 6, "contexts_frame_id": ["1229", "638", "1296", "96", "456", "346"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000331", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0328_00", "gt_frame_id": "1229", "anchor_frame_id": "1256", "shot": 6, "contexts_frame_id": ["1256", "656", "170", "108", "82", "858"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000334", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0329_00", "gt_frame_id": "357", "anchor_frame_id": "408", "shot": 6, "contexts_frame_id": ["408", "392", "1514", "1522", "512", "312"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000337", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0338_00", "gt_frame_id": "204", "anchor_frame_id": "237", "shot": 6, "contexts_frame_id": ["237", "800", "774", "172", "684", "124"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000340", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0338_01", "gt_frame_id": "238", "anchor_frame_id": "182", "shot": 6, "contexts_frame_id": ["182", "138", "654", "40", "756", "270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000343", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0338_01", "gt_frame_id": "249", "anchor_frame_id": "344", "shot": 6, "contexts_frame_id": ["344", "324", "296", "756", "378", "404"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000346", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0343_00", "gt_frame_id": "1178", "anchor_frame_id": "1079", "shot": 6, "contexts_frame_id": ["1079", "174", "216", "276", "400", "518"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000349", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "82", "anchor_frame_id": "321", "shot": 6, "contexts_frame_id": ["321", "1662", "1586", "2916", "2874", "1250"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000352", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "684", "anchor_frame_id": "548", "shot": 6, "contexts_frame_id": ["548", "644", "728", "794", "888", "2414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000355", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "794", "anchor_frame_id": "982", "shot": 6, "contexts_frame_id": ["982", "2662", "870", "842", "2344", "706"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.3 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000358", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "426", "anchor_frame_id": "680", "shot": 6, "contexts_frame_id": ["680", "2172", "1984", "522", "1722", "1696"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000361", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "989", "anchor_frame_id": "798", "shot": 6, "contexts_frame_id": ["798", "834", "866", "2668", "2450", "2162"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move right about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000364", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_00", "gt_frame_id": "317", "anchor_frame_id": "74", "shot": 6, "contexts_frame_id": ["74", "378", "1674", "1698", "408", "2926"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000367", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0353_01", "gt_frame_id": "103", "anchor_frame_id": "355", "shot": 6, "contexts_frame_id": ["355", "320", "1682", "2306", "1546", "378"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000370", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0353_01", "gt_frame_id": "1319", "anchor_frame_id": "1176", "shot": 6, "contexts_frame_id": ["1176", "1088", "1032", "1274", "1262", "1354"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000373", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0353_02", "gt_frame_id": "1258", "anchor_frame_id": "1468", "shot": 6, "contexts_frame_id": ["1468", "174", "1380", "2100", "2114", "260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000376", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0353_02", "gt_frame_id": "732", "anchor_frame_id": "623", "shot": 6, "contexts_frame_id": ["623", "1706", "568", "758", "1940", "1876"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000379", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0356_01", "gt_frame_id": "930", "anchor_frame_id": "994", "shot": 6, "contexts_frame_id": ["994", "394", "370", "330", "282", "762"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000382", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0356_01", "gt_frame_id": "917", "anchor_frame_id": "840", "shot": 6, "contexts_frame_id": ["840", "316", "320", "364", "758", "708"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000385", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0377_00", "gt_frame_id": "114", "anchor_frame_id": "53", "shot": 6, "contexts_frame_id": ["53", "810", "732", "566", "534", "372"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000388", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0377_02", "gt_frame_id": "605", "anchor_frame_id": "578", "shot": 6, "contexts_frame_id": ["578", "64", "500", "450", "646", "758"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000394", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0378_00", "gt_frame_id": "626", "anchor_frame_id": "407", "shot": 6, "contexts_frame_id": ["407", "432", "498", "822", "938", "696"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000397", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0378_01", "gt_frame_id": "1632", "anchor_frame_id": "1774", "shot": 6, "contexts_frame_id": ["1774", "1724", "1582", "1386", "1250", "1236"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000400", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0378_01", "gt_frame_id": "718", "anchor_frame_id": "519", "shot": 6, "contexts_frame_id": ["519", "480", "774", "892", "926", "1036"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000403", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0378_01", "gt_frame_id": "420", "anchor_frame_id": "623", "shot": 6, "contexts_frame_id": ["623", "2170", "2152", "576", "448", "390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000406", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0378_02", "gt_frame_id": "635", "anchor_frame_id": "586", "shot": 6, "contexts_frame_id": ["586", "542", "384", "482", "668", "942"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000409", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0378_02", "gt_frame_id": "555", "anchor_frame_id": "637", "shot": 6, "contexts_frame_id": ["637", "486", "580", "528", "750", "792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000412", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0378_02", "gt_frame_id": "1637", "anchor_frame_id": "1339", "shot": 6, "contexts_frame_id": ["1339", "1094", "1038", "1376", "1736", "1570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000415", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0426_00", "gt_frame_id": "1333", "anchor_frame_id": "1052", "shot": 6, "contexts_frame_id": ["1052", "1190", "1300", "1436", "52", "42"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000418", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0426_01", "gt_frame_id": "850", "anchor_frame_id": "1012", "shot": 6, "contexts_frame_id": ["1012", "1490", "1440", "1172", "1250", "1314"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move forward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000421", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0426_03", "gt_frame_id": "953", "anchor_frame_id": "1087", "shot": 6, "contexts_frame_id": ["1087", "1058", "922", "1050", "40", "86"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000424", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0430_00", "gt_frame_id": "1859", "anchor_frame_id": "1894", "shot": 6, "contexts_frame_id": ["1894", "1674", "1630", "1246", "2078", "1058"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000430", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0430_01", "gt_frame_id": "325", "anchor_frame_id": "198", "shot": 6, "contexts_frame_id": ["198", "66", "242", "356", "442", "538"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000439", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_01", "gt_frame_id": "1086", "anchor_frame_id": "894", "shot": 6, "contexts_frame_id": ["894", "1046", "830", "1182", "1194", "1266"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000442", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_01", "gt_frame_id": "906", "anchor_frame_id": "1076", "shot": 6, "contexts_frame_id": ["1076", "1172", "664", "760", "866", "972"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000445", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_02", "gt_frame_id": "249", "anchor_frame_id": "29", "shot": 6, "contexts_frame_id": ["29", "206", "1532", "1608", "1592", "1400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000448", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_02", "gt_frame_id": "37", "anchor_frame_id": "247", "shot": 6, "contexts_frame_id": ["247", "366", "404", "504", "82", "1614"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000451", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_02", "gt_frame_id": "867", "anchor_frame_id": "1038", "shot": 6, "contexts_frame_id": ["1038", "1078", "1134", "648", "588", "910"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000454", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0435_03", "gt_frame_id": "1037", "anchor_frame_id": "1010", "shot": 6, "contexts_frame_id": ["1010", "222", "1752", "978", "378", "454"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000457", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_03", "gt_frame_id": "1715", "anchor_frame_id": "1754", "shot": 6, "contexts_frame_id": ["1754", "1070", "1216", "1800", "162", "308"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000460", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0435_03", "gt_frame_id": "1757", "anchor_frame_id": "1715", "shot": 6, "contexts_frame_id": ["1715", "1098", "1078", "1788", "140", "310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000466", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0458_00", "gt_frame_id": "1044", "anchor_frame_id": "772", "shot": 6, "contexts_frame_id": ["772", "798", "1016", "212", "1374", "288"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000472", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0458_00", "gt_frame_id": "222", "anchor_frame_id": "306", "shot": 6, "contexts_frame_id": ["306", "1392", "1360", "1338", "1436", "1536"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000475", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0458_01", "gt_frame_id": "950", "anchor_frame_id": "1112", "shot": 6, "contexts_frame_id": ["1112", "742", "692", "1390", "1420", "1648"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000478", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0462_00", "gt_frame_id": "167", "anchor_frame_id": "304", "shot": 6, "contexts_frame_id": ["304", "1118", "120", "86", "894", "4"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000481", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0474_01", "gt_frame_id": "1324", "anchor_frame_id": "1077", "shot": 6, "contexts_frame_id": ["1077", "1180", "1414", "1378", "1494", "978"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000484", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0474_01", "gt_frame_id": "1078", "anchor_frame_id": "1318", "shot": 6, "contexts_frame_id": ["1318", "1174", "1184", "1024", "954", "642"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000493", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0474_02", "gt_frame_id": "1018", "anchor_frame_id": "840", "shot": 6, "contexts_frame_id": ["840", "1122", "1146", "1198", "1324", "1554"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then move left about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000496", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0474_02", "gt_frame_id": "846", "anchor_frame_id": "548", "shot": 6, "contexts_frame_id": ["548", "754", "998", "1150", "1198", "1324"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000502", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0474_02", "gt_frame_id": "1423", "anchor_frame_id": "1163", "shot": 6, "contexts_frame_id": ["1163", "1186", "1256", "1458", "1034", "922"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000505", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0474_02", "gt_frame_id": "140", "anchor_frame_id": "95", "shot": 6, "contexts_frame_id": ["95", "16", "190", "368", "564", "540"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000511", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0474_02", "gt_frame_id": "97", "anchor_frame_id": "141", "shot": 6, "contexts_frame_id": ["141", "2340", "36", "2184", "2146", "1928"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000514", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0488_00", "gt_frame_id": "89", "anchor_frame_id": "383", "shot": 6, "contexts_frame_id": ["383", "138", "208", "402", "40", "516"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.1 meters, then move backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000517", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0488_00", "gt_frame_id": "89", "anchor_frame_id": "185", "shot": 6, "contexts_frame_id": ["185", "138", "386", "402", "38", "516"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.3 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000520", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0488_00", "gt_frame_id": "156", "anchor_frame_id": "449", "shot": 6, "contexts_frame_id": ["449", "548", "102", "362", "264", "210"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees, then move left about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000523", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0488_00", "gt_frame_id": "376", "anchor_frame_id": "89", "shot": 6, "contexts_frame_id": ["89", "116", "156", "198", "536", "468"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then move forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000529", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0488_01", "gt_frame_id": "228", "anchor_frame_id": "53", "shot": 6, "contexts_frame_id": ["53", "120", "614", "280", "524", "428"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000535", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0488_01", "gt_frame_id": "348", "anchor_frame_id": "479", "shot": 6, "contexts_frame_id": ["479", "388", "268", "618", "650", "1088"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000538", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0488_01", "gt_frame_id": "1010", "anchor_frame_id": "1094", "shot": 6, "contexts_frame_id": ["1094", "834", "234", "334", "404", "482"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000541", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0490_00", "gt_frame_id": "727", "anchor_frame_id": "670", "shot": 6, "contexts_frame_id": ["670", "690", "1250", "1262", "1266", "150"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000550", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0500_01", "gt_frame_id": "489", "anchor_frame_id": "446", "shot": 6, "contexts_frame_id": ["446", "800", "836", "768", "564", "600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000553", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0500_01", "gt_frame_id": "227", "anchor_frame_id": "114", "shot": 6, "contexts_frame_id": ["114", "0", "132", "270", "310", "390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000556", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0527_00", "gt_frame_id": "87", "anchor_frame_id": "27", "shot": 6, "contexts_frame_id": ["27", "614", "642", "122", "184", "236"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000565", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0552_00", "gt_frame_id": "217", "anchor_frame_id": "290", "shot": 6, "contexts_frame_id": ["290", "258", "820", "800", "762", "176"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000568", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0552_00", "gt_frame_id": "197", "anchor_frame_id": "326", "shot": 6, "contexts_frame_id": ["326", "868", "238", "800", "732", "156"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000574", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0552_01", "gt_frame_id": "393", "anchor_frame_id": "357", "shot": 6, "contexts_frame_id": ["357", "332", "222", "116", "76", "56"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000580", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0552_01", "gt_frame_id": "242", "anchor_frame_id": "175", "shot": 6, "contexts_frame_id": ["175", "344", "214", "318", "296", "270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000583", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0552_01", "gt_frame_id": "353", "anchor_frame_id": "388", "shot": 6, "contexts_frame_id": ["388", "56", "164", "206", "326", "308"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000586", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0553_00", "gt_frame_id": "702", "anchor_frame_id": "803", "shot": 6, "contexts_frame_id": ["803", "630", "420", "982", "1090", "1452"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000589", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0553_00", "gt_frame_id": "803", "anchor_frame_id": "702", "shot": 6, "contexts_frame_id": ["702", "666", "882", "418", "354", "318"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000592", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0553_01", "gt_frame_id": "36", "anchor_frame_id": "250", "shot": 6, "contexts_frame_id": ["250", "360", "192", "106", "1276", "1198"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000598", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0553_02", "gt_frame_id": "491", "anchor_frame_id": "773", "shot": 6, "contexts_frame_id": ["773", "528", "630", "936", "1094", "1220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000601", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0568_01", "gt_frame_id": "409", "anchor_frame_id": "531", "shot": 6, "contexts_frame_id": ["531", "354", "328", "302", "236", "160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000604", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0568_02", "gt_frame_id": "389", "anchor_frame_id": "622", "shot": 6, "contexts_frame_id": ["622", "588", "496", "346", "662", "794"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000607", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0574_02", "gt_frame_id": "566", "anchor_frame_id": "300", "shot": 6, "contexts_frame_id": ["300", "1498", "1470", "934", "62", "1596"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move left about 1.2 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000610", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0575_00", "gt_frame_id": "1934", "anchor_frame_id": "1812", "shot": 6, "contexts_frame_id": ["1812", "54", "1088", "1878", "2008", "982"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000613", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0575_00", "gt_frame_id": "1749", "anchor_frame_id": "1941", "shot": 6, "contexts_frame_id": ["1941", "1146", "1238", "1314", "1570", "1634"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000619", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0575_00", "gt_frame_id": "1941", "anchor_frame_id": "1746", "shot": 6, "contexts_frame_id": ["1746", "1428", "1238", "1154", "1088", "1986"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000622", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0575_01", "gt_frame_id": "3027", "anchor_frame_id": "2993", "shot": 6, "contexts_frame_id": ["2993", "1624", "1778", "1930", "1472", "1280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000625", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0575_02", "gt_frame_id": "2730", "anchor_frame_id": "2964", "shot": 6, "contexts_frame_id": ["2964", "2984", "1624", "2784", "980", "2638"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000628", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0578_00", "gt_frame_id": "1189", "anchor_frame_id": "1074", "shot": 6, "contexts_frame_id": ["1074", "810", "786", "1138", "454", "1316"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000631", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0583_00", "gt_frame_id": "608", "anchor_frame_id": "497", "shot": 6, "contexts_frame_id": ["497", "462", "194", "222", "1746", "1576"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000634", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0583_01", "gt_frame_id": "715", "anchor_frame_id": "451", "shot": 6, "contexts_frame_id": ["451", "414", "368", "486", "1260", "1236"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000637", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0583_02", "gt_frame_id": "242", "anchor_frame_id": "182", "shot": 6, "contexts_frame_id": ["182", "944", "1100", "1272", "1508", "1484"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000640", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0591_00", "gt_frame_id": "1472", "anchor_frame_id": "1430", "shot": 6, "contexts_frame_id": ["1430", "22", "232", "292", "350", "1502"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000643", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0591_01", "gt_frame_id": "917", "anchor_frame_id": "783", "shot": 6, "contexts_frame_id": ["783", "826", "1576", "1698", "1538", "1510"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000646", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0591_01", "gt_frame_id": "779", "anchor_frame_id": "924", "shot": 6, "contexts_frame_id": ["924", "810", "988", "1064", "746", "252"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000649", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0591_02", "gt_frame_id": "407", "anchor_frame_id": "271", "shot": 6, "contexts_frame_id": ["271", "306", "366", "1370", "224", "2050"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000652", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0591_02", "gt_frame_id": "259", "anchor_frame_id": "413", "shot": 6, "contexts_frame_id": ["413", "1650", "286", "230", "610", "1406"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000655", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0591_02", "gt_frame_id": "189", "anchor_frame_id": "3", "shot": 6, "contexts_frame_id": ["3", "2186", "2274", "842", "1716", "944"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000658", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0591_02", "gt_frame_id": "1", "anchor_frame_id": "190", "shot": 6, "contexts_frame_id": ["190", "244", "810", "2238", "2204", "84"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000661", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0593_00", "gt_frame_id": "1245", "anchor_frame_id": "1103", "shot": 6, "contexts_frame_id": ["1103", "1004", "870", "1608", "1758", "1280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000676", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0593_01", "gt_frame_id": "1168", "anchor_frame_id": "910", "shot": 6, "contexts_frame_id": ["910", "606", "516", "504", "556", "844"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000679", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0595_00", "gt_frame_id": "676", "anchor_frame_id": "733", "shot": 6, "contexts_frame_id": ["733", "708", "36", "66", "110", "148"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000682", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0598_00", "gt_frame_id": "470", "anchor_frame_id": "395", "shot": 6, "contexts_frame_id": ["395", "438", "558", "288", "264", "204"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000685", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0598_00", "gt_frame_id": "396", "anchor_frame_id": "475", "shot": 6, "contexts_frame_id": ["475", "438", "322", "288", "264", "602"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000688", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0598_01", "gt_frame_id": "302", "anchor_frame_id": "400", "shot": 6, "contexts_frame_id": ["400", "430", "348", "216", "166", "786"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.6 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000691", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0598_01", "gt_frame_id": "398", "anchor_frame_id": "308", "shot": 6, "contexts_frame_id": ["308", "342", "192", "168", "786", "634"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 1.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000694", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0598_02", "gt_frame_id": "1477", "anchor_frame_id": "1548", "shot": 6, "contexts_frame_id": ["1548", "1568", "1606", "474", "552", "0"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then move right about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000697", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0598_02", "gt_frame_id": "1551", "anchor_frame_id": "1475", "shot": 6, "contexts_frame_id": ["1475", "480", "552", "0", "724", "178"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000700", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0599_02", "gt_frame_id": "155", "anchor_frame_id": "95", "shot": 6, "contexts_frame_id": ["95", "836", "210", "2388", "636", "1232"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000703", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0599_02", "gt_frame_id": "394", "anchor_frame_id": "486", "shot": 6, "contexts_frame_id": ["486", "1996", "432", "1030", "886", "1662"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000706", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0599_02", "gt_frame_id": "94", "anchor_frame_id": "155", "shot": 6, "contexts_frame_id": ["155", "876", "1572", "784", "2102", "2070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000709", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0599_02", "gt_frame_id": "1554", "anchor_frame_id": "1585", "shot": 6, "contexts_frame_id": ["1585", "1626", "372", "1714", "818", "800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000715", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0606_00", "gt_frame_id": "321", "anchor_frame_id": "124", "shot": 6, "contexts_frame_id": ["124", "166", "210", "456", "486", "546"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000718", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0606_01", "gt_frame_id": "443", "anchor_frame_id": "694", "shot": 6, "contexts_frame_id": ["694", "660", "566", "508", "338", "2070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000721", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0606_01", "gt_frame_id": "879", "anchor_frame_id": "935", "shot": 6, "contexts_frame_id": ["935", "922", "978", "800", "612", "676"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000724", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0606_02", "gt_frame_id": "829", "anchor_frame_id": "1088", "shot": 6, "contexts_frame_id": ["1088", "878", "734", "1128", "1188", "1242"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000727", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0607_00", "gt_frame_id": "837", "anchor_frame_id": "753", "shot": 6, "contexts_frame_id": ["753", "346", "204", "166", "450", "432"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000730", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0607_00", "gt_frame_id": "737", "anchor_frame_id": "840", "shot": 6, "contexts_frame_id": ["840", "464", "202", "232", "340", "694"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000736", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0607_01", "gt_frame_id": "242", "anchor_frame_id": "210", "shot": 6, "contexts_frame_id": ["210", "568", "538", "152", "80", "22"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000739", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0608_01", "gt_frame_id": "834", "anchor_frame_id": "732", "shot": 6, "contexts_frame_id": ["732", "768", "894", "426", "310", "262"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000742", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0608_01", "gt_frame_id": "719", "anchor_frame_id": "893", "shot": 6, "contexts_frame_id": ["893", "1032", "768", "478", "616", "2698"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000745", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0608_02", "gt_frame_id": "245", "anchor_frame_id": "155", "shot": 6, "contexts_frame_id": ["155", "210", "280", "366", "420", "2496"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000748", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0608_02", "gt_frame_id": "310", "anchor_frame_id": "432", "shot": 6, "contexts_frame_id": ["432", "364", "252", "872", "968", "1146"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000751", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0608_02", "gt_frame_id": "750", "anchor_frame_id": "899", "shot": 6, "contexts_frame_id": ["899", "2480", "384", "2164", "2614", "2684"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000754", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0616_00", "gt_frame_id": "653", "anchor_frame_id": "719", "shot": 6, "contexts_frame_id": ["719", "694", "2998", "2962", "2", "1616"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move right about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000757", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0616_00", "gt_frame_id": "1059", "anchor_frame_id": "1118", "shot": 6, "contexts_frame_id": ["1118", "1094", "1994", "114", "2232", "2844"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000760", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0616_00", "gt_frame_id": "124", "anchor_frame_id": "164", "shot": 6, "contexts_frame_id": ["164", "2820", "2736", "1032", "2210", "2188"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000763", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0616_00", "gt_frame_id": "90", "anchor_frame_id": "121", "shot": 6, "contexts_frame_id": ["121", "2658", "1994", "1100", "1940", "2896"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000766", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0616_01", "gt_frame_id": "759", "anchor_frame_id": "683", "shot": 6, "contexts_frame_id": ["683", "2316", "220", "242", "798", "2666"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000781", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0616_01", "gt_frame_id": "979", "anchor_frame_id": "1199", "shot": 6, "contexts_frame_id": ["1199", "3306", "1072", "318", "3078", "890"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000787", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0616_01", "gt_frame_id": "483", "anchor_frame_id": "447", "shot": 6, "contexts_frame_id": ["447", "1628", "3326", "130", "1842", "2292"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000793", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0621_00", "gt_frame_id": "1423", "anchor_frame_id": "1313", "shot": 6, "contexts_frame_id": ["1313", "2596", "928", "428", "486", "538"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000802", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "1379", "anchor_frame_id": "1624", "shot": 6, "contexts_frame_id": ["1624", "2618", "2580", "1318", "2360", "2332"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 1.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000805", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0629_00", "gt_frame_id": "327", "anchor_frame_id": "546", "shot": 6, "contexts_frame_id": ["546", "1986", "250", "402", "620", "720"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000808", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0629_00", "gt_frame_id": "2540", "anchor_frame_id": "2694", "shot": 6, "contexts_frame_id": ["2694", "1384", "1346", "2390", "2354", "138"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000811", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "2120", "anchor_frame_id": "1933", "shot": 6, "contexts_frame_id": ["1933", "102", "2000", "602", "718", "916"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000814", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0629_00", "gt_frame_id": "1771", "anchor_frame_id": "1528", "shot": 6, "contexts_frame_id": ["1528", "1488", "1388", "1360", "1298", "2360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then move left about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000817", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0629_00", "gt_frame_id": "334", "anchor_frame_id": "631", "shot": 6, "contexts_frame_id": ["631", "402", "1986", "250", "740", "924"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000823", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "2374", "anchor_frame_id": "2211", "shot": 6, "contexts_frame_id": ["2211", "934", "1120", "1094", "1208", "1326"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move backward about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000826", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "1624", "anchor_frame_id": "1373", "shot": 6, "contexts_frame_id": ["1373", "1304", "2592", "2646", "78", "232"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.9 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000829", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "2137", "anchor_frame_id": "2029", "shot": 6, "contexts_frame_id": ["2029", "486", "2090", "710", "2208", "922"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000835", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_00", "gt_frame_id": "1930", "anchor_frame_id": "2120", "shot": 6, "contexts_frame_id": ["2120", "592", "282", "250", "1470", "102"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000838", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "390", "anchor_frame_id": "561", "shot": 6, "contexts_frame_id": ["561", "2052", "230", "500", "672", "1172"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000841", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "2202", "anchor_frame_id": "2266", "shot": 6, "contexts_frame_id": ["2266", "734", "2150", "454", "2016", "1976"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000844", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "2264", "anchor_frame_id": "2228", "shot": 6, "contexts_frame_id": ["2228", "626", "526", "1100", "1274", "2300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000847", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "1337", "anchor_frame_id": "1403", "shot": 6, "contexts_frame_id": ["1403", "2564", "2446", "2380", "1296", "1268"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000850", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "1397", "anchor_frame_id": "1324", "shot": 6, "contexts_frame_id": ["1324", "2430", "1480", "1290", "1138", "2206"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000853", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "1320", "anchor_frame_id": "1224", "shot": 6, "contexts_frame_id": ["1224", "1138", "1280", "2360", "2400", "1374"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000856", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "558", "anchor_frame_id": "318", "shot": 6, "contexts_frame_id": ["318", "500", "2064", "1994", "222", "172"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000862", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0629_01", "gt_frame_id": "757", "anchor_frame_id": "997", "shot": 6, "contexts_frame_id": ["997", "2284", "948", "1210", "602", "500"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000865", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "1211", "anchor_frame_id": "1280", "shot": 6, "contexts_frame_id": ["1280", "988", "736", "2160", "454", "2016"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000868", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0629_01", "gt_frame_id": "2193", "anchor_frame_id": "1985", "shot": 6, "contexts_frame_id": ["1985", "500", "452", "2136", "1230", "818"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000871", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0629_02", "gt_frame_id": "338", "anchor_frame_id": "624", "shot": 6, "contexts_frame_id": ["624", "514", "268", "204", "716", "878"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000877", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0629_02", "gt_frame_id": "277", "anchor_frame_id": "335", "shot": 6, "contexts_frame_id": ["335", "360", "462", "540", "172", "102"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000880", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0629_02", "gt_frame_id": "1439", "anchor_frame_id": "1513", "shot": 6, "contexts_frame_id": ["1513", "1386", "56", "156", "246", "456"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then move backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000883", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0633_00", "gt_frame_id": "92", "anchor_frame_id": "190", "shot": 6, "contexts_frame_id": ["190", "128", "460", "1326", "16", "626"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000886", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0633_01", "gt_frame_id": "486", "anchor_frame_id": "575", "shot": 6, "contexts_frame_id": ["575", "802", "1340", "966", "1266", "262"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000889", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0643_00", "gt_frame_id": "449", "anchor_frame_id": "654", "shot": 6, "contexts_frame_id": ["654", "724", "1090", "556", "1134", "414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000892", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0644_00", "gt_frame_id": "909", "anchor_frame_id": "848", "shot": 6, "contexts_frame_id": ["848", "66", "114", "200", "244", "1306"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000895", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0644_00", "gt_frame_id": "694", "anchor_frame_id": "656", "shot": 6, "contexts_frame_id": ["656", "1536", "1476", "1374", "744", "1334"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000901", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0645_00", "gt_frame_id": "225", "anchor_frame_id": "10", "shot": 6, "contexts_frame_id": ["10", "176", "4440", "4420", "486", "528"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000904", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_00", "gt_frame_id": "225", "anchor_frame_id": "263", "shot": 6, "contexts_frame_id": ["263", "494", "540", "560", "4428", "2968"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000907", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0645_01", "gt_frame_id": "928", "anchor_frame_id": "723", "shot": 6, "contexts_frame_id": ["723", "996", "1534", "1086", "1162", "1756"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000910", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_01", "gt_frame_id": "456", "anchor_frame_id": "223", "shot": 6, "contexts_frame_id": ["223", "382", "502", "3534", "5172", "4990"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000913", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_01", "gt_frame_id": "197", "anchor_frame_id": "455", "shot": 6, "contexts_frame_id": ["455", "368", "576", "606", "240", "154"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000916", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0645_01", "gt_frame_id": "1727", "anchor_frame_id": "1772", "shot": 6, "contexts_frame_id": ["1772", "1314", "1162", "2902", "1900", "3066"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000919", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_02", "gt_frame_id": "2901", "anchor_frame_id": "2736", "shot": 6, "contexts_frame_id": ["2736", "2574", "2542", "2764", "150", "94"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000922", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_02", "gt_frame_id": "2835", "anchor_frame_id": "2591", "shot": 6, "contexts_frame_id": ["2591", "2766", "2810", "198", "220", "360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000925", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0645_02", "gt_frame_id": "160", "anchor_frame_id": "10", "shot": 6, "contexts_frame_id": ["10", "2792", "2660", "2700", "2556", "2490"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000928", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_02", "gt_frame_id": "2591", "anchor_frame_id": "2835", "shot": 6, "contexts_frame_id": ["2835", "2792", "2682", "2554", "48", "228"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move right about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000931", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0645_02", "gt_frame_id": "2739", "anchor_frame_id": "2835", "shot": 6, "contexts_frame_id": ["2835", "48", "160", "2780", "2662", "2586"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000934", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0647_01", "gt_frame_id": "779", "anchor_frame_id": "802", "shot": 6, "contexts_frame_id": ["802", "830", "916", "950", "4", "110"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000937", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0648_00", "gt_frame_id": "2600", "anchor_frame_id": "2734", "shot": 6, "contexts_frame_id": ["2734", "1176", "1232", "2296", "1306", "220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move left about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000940", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0648_00", "gt_frame_id": "746", "anchor_frame_id": "702", "shot": 6, "contexts_frame_id": ["702", "0", "48", "3846", "3256", "4086"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000943", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0648_01", "gt_frame_id": "1473", "anchor_frame_id": "1651", "shot": 6, "contexts_frame_id": ["1651", "2492", "2448", "896", "2420", "1366"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000946", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0651_00", "gt_frame_id": "359", "anchor_frame_id": "207", "shot": 6, "contexts_frame_id": ["207", "1022", "276", "324", "304", "394"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000949", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0651_01", "gt_frame_id": "330", "anchor_frame_id": "181", "shot": 6, "contexts_frame_id": ["181", "212", "278", "252", "380", "432"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.6 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000952", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0651_02", "gt_frame_id": "112", "anchor_frame_id": "205", "shot": 6, "contexts_frame_id": ["205", "746", "722", "2", "264", "312"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000955", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0652_00", "gt_frame_id": "463", "anchor_frame_id": "378", "shot": 6, "contexts_frame_id": ["378", "406", "524", "1216", "278", "164"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000958", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0653_00", "gt_frame_id": "3790", "anchor_frame_id": "3643", "shot": 6, "contexts_frame_id": ["3643", "3678", "358", "194", "382", "592"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000961", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0653_00", "gt_frame_id": "2172", "anchor_frame_id": "2243", "shot": 6, "contexts_frame_id": ["2243", "2264", "2094", "2382", "2508", "2576"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000964", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0653_01", "gt_frame_id": "3837", "anchor_frame_id": "3923", "shot": 6, "contexts_frame_id": ["3923", "576", "3790", "1368", "1396", "1428"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000985", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0653_01", "gt_frame_id": "798", "anchor_frame_id": "952", "shot": 6, "contexts_frame_id": ["952", "1016", "586", "710", "3824", "3988"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then move backward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00000991", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0653_01", "gt_frame_id": "4740", "anchor_frame_id": "4637", "shot": 6, "contexts_frame_id": ["4637", "2938", "3190", "90", "3240", "440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001000", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0655_00", "gt_frame_id": "162", "anchor_frame_id": "307", "shot": 6, "contexts_frame_id": ["307", "244", "636", "604", "588", "566"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move right about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001003", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0655_00", "gt_frame_id": "652", "anchor_frame_id": "575", "shot": 6, "contexts_frame_id": ["575", "30", "178", "356", "224", "330"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.0 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001009", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0655_01", "gt_frame_id": "528", "anchor_frame_id": "509", "shot": 6, "contexts_frame_id": ["509", "360", "2", "48", "130", "190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001012", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0655_02", "gt_frame_id": "140", "anchor_frame_id": "183", "shot": 6, "contexts_frame_id": ["183", "206", "532", "480", "362", "6"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001015", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0663_00", "gt_frame_id": "494", "anchor_frame_id": "575", "shot": 6, "contexts_frame_id": ["575", "554", "84", "1538", "1504", "1324"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001018", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0663_00", "gt_frame_id": "555", "anchor_frame_id": "755", "shot": 6, "contexts_frame_id": ["755", "1818", "676", "590", "492", "2922"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001021", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0663_00", "gt_frame_id": "644", "anchor_frame_id": "755", "shot": 6, "contexts_frame_id": ["755", "1770", "600", "568", "1818", "1056"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001024", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0663_01", "gt_frame_id": "1068", "anchor_frame_id": "981", "shot": 6, "contexts_frame_id": ["981", "664", "1212", "1420", "1298", "1498"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001030", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0663_01", "gt_frame_id": "2031", "anchor_frame_id": "1959", "shot": 6, "contexts_frame_id": ["1959", "1638", "1514", "2352", "3106", "2250"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001036", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0663_01", "gt_frame_id": "2484", "anchor_frame_id": "2454", "shot": 6, "contexts_frame_id": ["2454", "2298", "2154", "116", "1908", "1848"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001045", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0663_02", "gt_frame_id": "1250", "anchor_frame_id": "990", "shot": 6, "contexts_frame_id": ["990", "1052", "792", "800", "1410", "1446"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001051", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0670_00", "gt_frame_id": "716", "anchor_frame_id": "972", "shot": 6, "contexts_frame_id": ["972", "902", "660", "3026", "2798", "2732"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001054", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0670_00", "gt_frame_id": "2622", "anchor_frame_id": "2598", "shot": 6, "contexts_frame_id": ["2598", "2220", "130", "344", "392", "2810"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001057", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0670_00", "gt_frame_id": "1611", "anchor_frame_id": "1633", "shot": 6, "contexts_frame_id": ["1633", "1498", "1890", "2422", "1992", "2546"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001063", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_00", "gt_frame_id": "560", "anchor_frame_id": "344", "shot": 6, "contexts_frame_id": ["344", "134", "2716", "520", "612", "754"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001066", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_00", "gt_frame_id": "150", "anchor_frame_id": "344", "shot": 6, "contexts_frame_id": ["344", "392", "2810", "2852", "2150", "86"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001069", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0670_00", "gt_frame_id": "998", "anchor_frame_id": "716", "shot": 6, "contexts_frame_id": ["716", "782", "1040", "1138", "1260", "1350"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001072", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_00", "gt_frame_id": "820", "anchor_frame_id": "563", "shot": 6, "contexts_frame_id": ["563", "722", "942", "2788", "2716", "2676"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001075", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_00", "gt_frame_id": "493", "anchor_frame_id": "563", "shot": 6, "contexts_frame_id": ["563", "2812", "2756", "730", "828", "1080"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001078", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0670_00", "gt_frame_id": "2688", "anchor_frame_id": "2749", "shot": 6, "contexts_frame_id": ["2749", "528", "3022", "764", "320", "2648"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001081", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0670_01", "gt_frame_id": "1856", "anchor_frame_id": "2142", "shot": 6, "contexts_frame_id": ["2142", "2128", "2232", "2344", "2402", "2572"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001087", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_01", "gt_frame_id": "3287", "anchor_frame_id": "3473", "shot": 6, "contexts_frame_id": ["3473", "3372", "3394", "3168", "450", "362"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move right about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001090", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0670_01", "gt_frame_id": "1439", "anchor_frame_id": "1641", "shot": 6, "contexts_frame_id": ["1641", "1668", "1726", "2226", "1490", "1292"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001093", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0670_01", "gt_frame_id": "3475", "anchor_frame_id": "3288", "shot": 6, "contexts_frame_id": ["3288", "3114", "3400", "34", "154", "130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.3 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001096", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0678_00", "gt_frame_id": "2169", "anchor_frame_id": "2111", "shot": 6, "contexts_frame_id": ["2111", "2090", "984", "2204", "202", "158"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001099", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0678_01", "gt_frame_id": "866", "anchor_frame_id": "843", "shot": 6, "contexts_frame_id": ["843", "914", "2028", "1936", "730", "686"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001102", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_00", "gt_frame_id": "1720", "anchor_frame_id": "1872", "shot": 6, "contexts_frame_id": ["1872", "126", "1796", "1676", "1632", "1418"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001105", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_00", "gt_frame_id": "1872", "anchor_frame_id": "1720", "shot": 6, "contexts_frame_id": ["1720", "1752", "1956", "1622", "1450", "1398"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001108", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_01", "gt_frame_id": "483", "anchor_frame_id": "517", "shot": 6, "contexts_frame_id": ["517", "378", "568", "626", "914", "968"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001111", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_01", "gt_frame_id": "493", "anchor_frame_id": "786", "shot": 6, "contexts_frame_id": ["786", "558", "598", "430", "388", "198"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001114", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_02", "gt_frame_id": "2473", "anchor_frame_id": "2650", "shot": 6, "contexts_frame_id": ["2650", "3182", "420", "2298", "334", "2100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.8 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001117", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0685_02", "gt_frame_id": "1121", "anchor_frame_id": "914", "shot": 6, "contexts_frame_id": ["914", "2960", "864", "1000", "1170", "1366"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then move backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001120", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0685_02", "gt_frame_id": "203", "anchor_frame_id": "336", "shot": 6, "contexts_frame_id": ["336", "2492", "2298", "2276", "254", "2100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001123", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0685_02", "gt_frame_id": "1536", "anchor_frame_id": "1369", "shot": 6, "contexts_frame_id": ["1369", "1476", "2098", "328", "2318", "2490"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001126", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0685_02", "gt_frame_id": "530", "anchor_frame_id": "726", "shot": 6, "contexts_frame_id": ["726", "2554", "502", "484", "2472", "3232"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001129", "dataset": "scannet", "task_type": "translation", "scene_id": "scene0685_02", "gt_frame_id": "2126", "anchor_frame_id": "2112", "shot": 6, "contexts_frame_id": ["2112", "1490", "328", "2318", "2490", "2546"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001132", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0685_02", "gt_frame_id": "920", "anchor_frame_id": "986", "shot": 6, "contexts_frame_id": ["986", "956", "1108", "2970", "3000", "2536"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001135", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0685_02", "gt_frame_id": "1334", "anchor_frame_id": "1607", "shot": 6, "contexts_frame_id": ["1607", "2126", "1378", "1240", "1114", "2276"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001138", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0685_02", "gt_frame_id": "332", "anchor_frame_id": "204", "shot": 6, "contexts_frame_id": ["204", "70", "2086", "3252", "370", "2494"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001141", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0693_01", "gt_frame_id": "529", "anchor_frame_id": "813", "shot": 6, "contexts_frame_id": ["813", "1014", "744", "470", "30", "100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001144", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0693_01", "gt_frame_id": "813", "anchor_frame_id": "529", "shot": 6, "contexts_frame_id": ["529", "408", "696", "862", "1000", "1078"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001147", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0693_02", "gt_frame_id": "379", "anchor_frame_id": "670", "shot": 6, "contexts_frame_id": ["670", "816", "586", "222", "52", "1590"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001153", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0695_00", "gt_frame_id": "1116", "anchor_frame_id": "1061", "shot": 6, "contexts_frame_id": ["1061", "142", "192", "1040", "1008", "702"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001156", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0695_01", "gt_frame_id": "2097", "anchor_frame_id": "2064", "shot": 6, "contexts_frame_id": ["2064", "1470", "1788", "1812", "536", "806"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001159", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0695_02", "gt_frame_id": "179", "anchor_frame_id": "240", "shot": 6, "contexts_frame_id": ["240", "942", "1008", "444", "1356", "1752"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001165", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0695_03", "gt_frame_id": "2486", "anchor_frame_id": "2608", "shot": 6, "contexts_frame_id": ["2608", "2582", "2392", "2436", "46", "32"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001168", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0695_03", "gt_frame_id": "2370", "anchor_frame_id": "2405", "shot": 6, "contexts_frame_id": ["2405", "2234", "422", "2604", "260", "1896"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001177", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0696_00", "gt_frame_id": "59", "anchor_frame_id": "103", "shot": 6, "contexts_frame_id": ["103", "848", "758", "418", "1142", "1194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001180", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0696_01", "gt_frame_id": "1499", "anchor_frame_id": "1447", "shot": 6, "contexts_frame_id": ["1447", "152", "854", "88", "666", "1286"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001183", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0696_01", "gt_frame_id": "1457", "anchor_frame_id": "1496", "shot": 6, "contexts_frame_id": ["1496", "166", "1414", "110", "678", "1288"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001186", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0696_02", "gt_frame_id": "1121", "anchor_frame_id": "1310", "shot": 6, "contexts_frame_id": ["1310", "1290", "1438", "1468", "1000", "790"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001192", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0697_02", "gt_frame_id": "1220", "anchor_frame_id": "1059", "shot": 6, "contexts_frame_id": ["1059", "634", "1156", "1318", "1406", "1674"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001195", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0697_02", "gt_frame_id": "1051", "anchor_frame_id": "1215", "shot": 6, "contexts_frame_id": ["1215", "1806", "1122", "2064", "1944", "888"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move backward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001198", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0697_02", "gt_frame_id": "868", "anchor_frame_id": "762", "shot": 6, "contexts_frame_id": ["762", "824", "2108", "1000", "1572", "1600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001201", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0697_03", "gt_frame_id": "195", "anchor_frame_id": "84", "shot": 6, "contexts_frame_id": ["84", "1128", "250", "318", "668", "2604"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001204", "dataset": "scannet", "task_type": "composition", "scene_id": "scene0697_03", "gt_frame_id": "69", "anchor_frame_id": "197", "shot": 6, "contexts_frame_id": ["197", "1144", "2674", "2604", "1940", "276"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001207", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0697_03", "gt_frame_id": "419", "anchor_frame_id": "468", "shot": 6, "contexts_frame_id": ["468", "2388", "2218", "1432", "1276", "1220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001210", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0697_03", "gt_frame_id": "467", "anchor_frame_id": "419", "shot": 6, "contexts_frame_id": ["419", "368", "318", "1544", "2422", "512"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001213", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0700_01", "gt_frame_id": "449", "anchor_frame_id": "501", "shot": 6, "contexts_frame_id": ["501", "1216", "1472", "2284", "2466", "2168"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001219", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0704_00", "gt_frame_id": "390", "anchor_frame_id": "547", "shot": 6, "contexts_frame_id": ["547", "846", "860", "1440", "234", "484"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001222", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0704_00", "gt_frame_id": "510", "anchor_frame_id": "404", "shot": 6, "contexts_frame_id": ["404", "860", "846", "546", "1010", "272"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001225", "dataset": "scannet", "task_type": "rotation", "scene_id": "scene0704_00", "gt_frame_id": "262", "anchor_frame_id": "101", "shot": 6, "contexts_frame_id": ["101", "288", "236", "1734", "1292", "1266"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001228", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0015_00", "gt_frame_id": "1708", "anchor_frame_id": "827", "shot": 6, "contexts_frame_id": ["827", "790", "744", "1632", "584", "1552"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001231", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0030_02", "gt_frame_id": "1914", "anchor_frame_id": "1178", "shot": 6, "contexts_frame_id": ["1178", "1866", "1082", "1034", "982", "78"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001234", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0030_02", "gt_frame_id": "1178", "anchor_frame_id": "1895", "shot": 6, "contexts_frame_id": ["1895", "1134", "204", "130", "1340", "1382"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001237", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0050_00", "gt_frame_id": "890", "anchor_frame_id": "828", "shot": 6, "contexts_frame_id": ["828", "1128", "1190", "1314", "1278", "444"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the piano, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the piano, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001240", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0081_00", "gt_frame_id": "82", "anchor_frame_id": "226", "shot": 6, "contexts_frame_id": ["226", "202", "122", "154", "254", "294"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001243", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0081_00", "gt_frame_id": "226", "anchor_frame_id": "82", "shot": 6, "contexts_frame_id": ["82", "68", "38", "294", "102", "124"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001246", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0088_02", "gt_frame_id": "750", "anchor_frame_id": "1892", "shot": 6, "contexts_frame_id": ["1892", "1360", "1848", "1784", "926", "1084"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the board, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the board, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001249", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0144_00", "gt_frame_id": "1424", "anchor_frame_id": "720", "shot": 6, "contexts_frame_id": ["720", "690", "914", "124", "1336", "1092"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001252", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0144_00", "gt_frame_id": "720", "anchor_frame_id": "1424", "shot": 6, "contexts_frame_id": ["1424", "658", "1306", "174", "570", "1006"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001255", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0144_01", "gt_frame_id": "1002", "anchor_frame_id": "225", "shot": 6, "contexts_frame_id": ["225", "1092", "274", "326", "360", "412"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001258", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "786", "anchor_frame_id": "662", "shot": 6, "contexts_frame_id": ["662", "706", "412", "2214", "2112", "2070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001261", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "662", "anchor_frame_id": "787", "shot": 6, "contexts_frame_id": ["787", "2158", "2084", "2030", "1230", "88"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001264", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "1381", "anchor_frame_id": "1563", "shot": 6, "contexts_frame_id": ["1563", "1600", "1712", "1774", "1306", "180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 90 degrees to the right around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001267", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "255", "anchor_frame_id": "1392", "shot": 6, "contexts_frame_id": ["1392", "1322", "1588", "1634", "1722", "1784"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 90 degrees to the right around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001270", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "1522", "anchor_frame_id": "19", "shot": 6, "contexts_frame_id": ["19", "1472", "1240", "1568", "1610", "290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001273", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_00", "gt_frame_id": "1920", "anchor_frame_id": "1075", "shot": 6, "contexts_frame_id": ["1075", "1980", "2014", "2070", "888", "784"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001276", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_01", "gt_frame_id": "132", "anchor_frame_id": "1672", "shot": 6, "contexts_frame_id": ["1672", "90", "1500", "274", "1742", "862"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001279", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0187_01", "gt_frame_id": "338", "anchor_frame_id": "403", "shot": 6, "contexts_frame_id": ["403", "282", "418", "822", "764", "916"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001282", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0203_00", "gt_frame_id": "1428", "anchor_frame_id": "856", "shot": 6, "contexts_frame_id": ["856", "1398", "930", "834", "814", "792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001285", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0203_00", "gt_frame_id": "860", "anchor_frame_id": "1428", "shot": 6, "contexts_frame_id": ["1428", "1396", "1298", "1282", "1352", "930"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001288", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0207_02", "gt_frame_id": "341", "anchor_frame_id": "959", "shot": 6, "contexts_frame_id": ["959", "928", "1032", "2238", "2420", "94"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the clothes, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the clothes, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001291", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0222_00", "gt_frame_id": "2410", "anchor_frame_id": "2879", "shot": 6, "contexts_frame_id": ["2879", "2446", "4948", "4986", "5418", "5130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the window, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the window, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001294", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0231_01", "gt_frame_id": "3284", "anchor_frame_id": "611", "shot": 6, "contexts_frame_id": ["611", "4146", "3254", "3530", "1714", "1844"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the kitchen cabinet, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the kitchen cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001297", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0328_00", "gt_frame_id": "159", "anchor_frame_id": "672", "shot": 6, "contexts_frame_id": ["672", "1190", "104", "88", "438", "306"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the window, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the window, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001300", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0329_01", "gt_frame_id": "275", "anchor_frame_id": "1097", "shot": 6, "contexts_frame_id": ["1097", "1084", "1050", "310", "366", "390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001303", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0334_00", "gt_frame_id": "1170", "anchor_frame_id": "125", "shot": 6, "contexts_frame_id": ["125", "210", "268", "1120", "1012", "966"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001306", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0334_00", "gt_frame_id": "1149", "anchor_frame_id": "125", "shot": 6, "contexts_frame_id": ["125", "210", "268", "950", "990", "1038"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001309", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0334_00", "gt_frame_id": "1122", "anchor_frame_id": "1182", "shot": 6, "contexts_frame_id": ["1182", "1044", "184", "262", "348", "610"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001312", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0353_00", "gt_frame_id": "1566", "anchor_frame_id": "158", "shot": 6, "contexts_frame_id": ["158", "1526", "2934", "228", "1674", "1702"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001315", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0355_00", "gt_frame_id": "226", "anchor_frame_id": "252", "shot": 6, "contexts_frame_id": ["252", "170", "22", "1016", "408", "420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001318", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0355_00", "gt_frame_id": "278", "anchor_frame_id": "252", "shot": 6, "contexts_frame_id": ["252", "90", "324", "398", "920", "420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001321", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0378_00", "gt_frame_id": "150", "anchor_frame_id": "1884", "shot": 6, "contexts_frame_id": ["1884", "186", "1710", "1742", "1358", "1278"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the desk, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the desk, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001324", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0378_00", "gt_frame_id": "873", "anchor_frame_id": "737", "shot": 6, "contexts_frame_id": ["737", "786", "486", "966", "1120", "1194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001327", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0378_00", "gt_frame_id": "731", "anchor_frame_id": "864", "shot": 6, "contexts_frame_id": ["864", "948", "1066", "786", "486", "442"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001330", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0378_01", "gt_frame_id": "732", "anchor_frame_id": "887", "shot": 6, "contexts_frame_id": ["887", "970", "1164", "1208", "774", "536"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001333", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0389_00", "gt_frame_id": "1002", "anchor_frame_id": "1043", "shot": 6, "contexts_frame_id": ["1043", "1134", "1238", "1304", "1326", "142"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the mini fridge, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the mini fridge, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001336", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0426_00", "gt_frame_id": "1626", "anchor_frame_id": "208", "shot": 6, "contexts_frame_id": ["208", "1586", "264", "352", "480", "592"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the jacket, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the jacket, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001339", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0427_00", "gt_frame_id": "925", "anchor_frame_id": "1091", "shot": 6, "contexts_frame_id": ["1091", "1114", "894", "346", "446", "510"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001342", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0430_00", "gt_frame_id": "2105", "anchor_frame_id": "2009", "shot": 6, "contexts_frame_id": ["2009", "1218", "2180", "1040", "1944", "1670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001345", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0435_02", "gt_frame_id": "366", "anchor_frame_id": "293", "shot": 6, "contexts_frame_id": ["293", "208", "392", "404", "504", "564"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the mini fridge, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the mini fridge, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001348", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0435_02", "gt_frame_id": "293", "anchor_frame_id": "366", "shot": 6, "contexts_frame_id": ["366", "344", "168", "412", "504", "564"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the mini fridge, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the mini fridge, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001351", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0527_00", "gt_frame_id": "21", "anchor_frame_id": "603", "shot": 6, "contexts_frame_id": ["603", "54", "96", "538", "464", "446"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bathroom vanity, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the bathroom vanity, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001354", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0527_00", "gt_frame_id": "603", "anchor_frame_id": "21", "shot": 6, "contexts_frame_id": ["21", "538", "642", "82", "152", "190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bathroom vanity, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the bathroom vanity, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001357", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0549_00", "gt_frame_id": "426", "anchor_frame_id": "455", "shot": 6, "contexts_frame_id": ["455", "312", "670", "678", "690", "768"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001360", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0553_00", "gt_frame_id": "1458", "anchor_frame_id": "1115", "shot": 6, "contexts_frame_id": ["1115", "992", "846", "648", "460", "1416"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sink, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the sink, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001363", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0553_00", "gt_frame_id": "1118", "anchor_frame_id": "1460", "shot": 6, "contexts_frame_id": ["1460", "1246", "1490", "190", "258", "318"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sink, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the sink, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001366", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0559_00", "gt_frame_id": "432", "anchor_frame_id": "400", "shot": 6, "contexts_frame_id": ["400", "358", "330", "258", "518", "574"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bench, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the bench, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001369", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0575_00", "gt_frame_id": "533", "anchor_frame_id": "934", "shot": 6, "contexts_frame_id": ["934", "982", "1028", "1102", "780", "490"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the office chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the office chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001372", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0575_00", "gt_frame_id": "934", "anchor_frame_id": "539", "shot": 6, "contexts_frame_id": ["539", "636", "2330", "336", "2682", "974"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the office chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the office chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001375", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0575_01", "gt_frame_id": "1509", "anchor_frame_id": "1555", "shot": 6, "contexts_frame_id": ["1555", "1940", "1460", "2874", "2914", "182"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001378", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0593_00", "gt_frame_id": "2116", "anchor_frame_id": "450", "shot": 6, "contexts_frame_id": ["450", "2010", "1942", "2158", "292", "222"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the trash can, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the trash can, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001381", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0593_00", "gt_frame_id": "450", "anchor_frame_id": "2116", "shot": 6, "contexts_frame_id": ["2116", "400", "1946", "2156", "298", "252"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the trash can, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the trash can, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001384", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0606_02", "gt_frame_id": "803", "anchor_frame_id": "864", "shot": 6, "contexts_frame_id": ["864", "742", "1056", "1118", "1176", "1230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the toaster oven, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the toaster oven, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001387", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0608_00", "gt_frame_id": "3196", "anchor_frame_id": "2147", "shot": 6, "contexts_frame_id": ["2147", "2282", "3306", "2372", "2662", "2624"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the guitar, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the guitar, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001390", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0608_02", "gt_frame_id": "2825", "anchor_frame_id": "101", "shot": 6, "contexts_frame_id": ["101", "2866", "220", "258", "1768", "2782"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the keyboard piano, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the keyboard piano, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001393", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0608_02", "gt_frame_id": "101", "anchor_frame_id": "2825", "shot": 6, "contexts_frame_id": ["2825", "2784", "1752", "2018", "1886", "170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the keyboard piano, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the keyboard piano, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001396", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0609_00", "gt_frame_id": "640", "anchor_frame_id": "1009", "shot": 6, "contexts_frame_id": ["1009", "786", "586", "510", "430", "166"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 90 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001399", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0609_03", "gt_frame_id": "447", "anchor_frame_id": "73", "shot": 6, "contexts_frame_id": ["73", "16", "594", "412", "372", "136"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001402", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "1456", "anchor_frame_id": "520", "shot": 6, "contexts_frame_id": ["520", "24", "3016", "296", "792", "912"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001405", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "2", "anchor_frame_id": "520", "shot": 6, "contexts_frame_id": ["520", "364", "1568", "2954", "296", "792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001408", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "585", "anchor_frame_id": "520", "shot": 6, "contexts_frame_id": ["520", "1574", "1908", "78", "2624", "3008"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001411", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "1121", "anchor_frame_id": "2101", "shot": 6, "contexts_frame_id": ["2101", "2130", "1684", "2896", "2948", "2980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the windowsill, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the windowsill, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001414", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "508", "anchor_frame_id": "1562", "shot": 6, "contexts_frame_id": ["1562", "1910", "78", "2624", "1056", "1380"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001417", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "1565", "anchor_frame_id": "505", "shot": 6, "contexts_frame_id": ["505", "3004", "2910", "32", "1934", "2190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001420", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_00", "gt_frame_id": "1247", "anchor_frame_id": "1225", "shot": 6, "contexts_frame_id": ["1225", "26", "1374", "1950", "90", "2538"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001423", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_01", "gt_frame_id": "1140", "anchor_frame_id": "1362", "shot": 6, "contexts_frame_id": ["1362", "3250", "3212", "1066", "318", "2640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001426", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_01", "gt_frame_id": "24", "anchor_frame_id": "1362", "shot": 6, "contexts_frame_id": ["1362", "330", "1072", "3216", "3246", "76"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 90 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001429", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_01", "gt_frame_id": "1915", "anchor_frame_id": "1978", "shot": 6, "contexts_frame_id": ["1978", "1830", "2064", "660", "536", "3328"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001432", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0616_01", "gt_frame_id": "2695", "anchor_frame_id": "2634", "shot": 6, "contexts_frame_id": ["2634", "2528", "3036", "3194", "1060", "316"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001435", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0629_02", "gt_frame_id": "905", "anchor_frame_id": "973", "shot": 6, "contexts_frame_id": ["973", "1122", "1042", "762", "686", "574"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001438", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0643_00", "gt_frame_id": "59", "anchor_frame_id": "206", "shot": 6, "contexts_frame_id": ["206", "152", "1356", "1162", "1124", "414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the desk, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the desk, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001441", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1297", "anchor_frame_id": "1528", "shot": 6, "contexts_frame_id": ["1528", "1448", "1184", "1016", "838", "670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the couch, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the couch, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001444", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1337", "anchor_frame_id": "1528", "shot": 6, "contexts_frame_id": ["1528", "1448", "1270", "1184", "1016", "838"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the couch, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the couch, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001447", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1488", "anchor_frame_id": "1297", "shot": 6, "contexts_frame_id": ["1297", "1436", "1646", "1158", "1012", "918"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the couch, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the couch, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001450", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "2105", "anchor_frame_id": "1951", "shot": 6, "contexts_frame_id": ["1951", "2046", "1766", "1642", "2570", "2364"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the desk, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the desk, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001453", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1457", "anchor_frame_id": "1409", "shot": 6, "contexts_frame_id": ["1409", "1500", "1222", "1152", "1010", "916"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001456", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1377", "anchor_frame_id": "1448", "shot": 6, "contexts_frame_id": ["1448", "1426", "1836", "1242", "1152", "1010"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001459", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_00", "gt_frame_id": "1417", "anchor_frame_id": "1377", "shot": 6, "contexts_frame_id": ["1377", "1460", "1636", "1952", "2570", "2608"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001462", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1044", "anchor_frame_id": "1608", "shot": 6, "contexts_frame_id": ["1608", "898", "1480", "1142", "1210", "1272"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001465", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1069", "anchor_frame_id": "1608", "shot": 6, "contexts_frame_id": ["1608", "898", "1480", "1142", "1210", "1284"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001468", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1608", "anchor_frame_id": "1069", "shot": 6, "contexts_frame_id": ["1069", "1530", "1452", "1670", "1772", "1862"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the armchair, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the armchair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001471", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1252", "anchor_frame_id": "1300", "shot": 6, "contexts_frame_id": ["1300", "1416", "1644", "1772", "2046", "2994"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001474", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1310", "anchor_frame_id": "1271", "shot": 6, "contexts_frame_id": ["1271", "2972", "1990", "1210", "1452", "1512"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001477", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0645_01", "gt_frame_id": "1187", "anchor_frame_id": "1644", "shot": 6, "contexts_frame_id": ["1644", "1502", "1142", "1752", "3276", "1252"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001480", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0651_02", "gt_frame_id": "572", "anchor_frame_id": "492", "shot": 6, "contexts_frame_id": ["492", "522", "610", "438", "402", "360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001483", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0651_02", "gt_frame_id": "595", "anchor_frame_id": "492", "shot": 6, "contexts_frame_id": ["492", "552", "636", "438", "402", "360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001486", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0652_00", "gt_frame_id": "807", "anchor_frame_id": "1", "shot": 6, "contexts_frame_id": ["1", "132", "102", "1364", "174", "332"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001489", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0652_00", "gt_frame_id": "15", "anchor_frame_id": "831", "shot": 6, "contexts_frame_id": ["831", "756", "1376", "160", "298", "378"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001492", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0653_01", "gt_frame_id": "2676", "anchor_frame_id": "4571", "shot": 6, "contexts_frame_id": ["4571", "4526", "2744", "2270", "4436", "4416"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the file cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the file cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001495", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0653_01", "gt_frame_id": "4571", "anchor_frame_id": "2676", "shot": 6, "contexts_frame_id": ["2676", "2904", "2928", "3024", "4678", "304"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the file cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the file cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001498", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0653_01", "gt_frame_id": "163", "anchor_frame_id": "3153", "shot": 6, "contexts_frame_id": ["3153", "2960", "4748", "6", "352", "450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the file cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the file cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001501", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0653_01", "gt_frame_id": "3165", "anchor_frame_id": "169", "shot": 6, "contexts_frame_id": ["169", "4668", "3216", "336", "3342", "3394"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the file cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the file cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001504", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0653_01", "gt_frame_id": "1511", "anchor_frame_id": "4406", "shot": 6, "contexts_frame_id": ["4406", "2138", "1542", "1780", "1638", "4256"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the office chair, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the office chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001507", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0663_00", "gt_frame_id": "323", "anchor_frame_id": "22", "shot": 6, "contexts_frame_id": ["22", "144", "1620", "466", "2926", "1426"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the office chair, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the office chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001510", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0663_01", "gt_frame_id": "1749", "anchor_frame_id": "2708", "shot": 6, "contexts_frame_id": ["2708", "2726", "1550", "2762", "2784", "2830"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001513", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0663_01", "gt_frame_id": "2718", "anchor_frame_id": "1752", "shot": 6, "contexts_frame_id": ["1752", "1820", "3072", "1538", "1498", "1424"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001516", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0663_02", "gt_frame_id": "2066", "anchor_frame_id": "698", "shot": 6, "contexts_frame_id": ["698", "2930", "202", "88", "2110", "2772"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001519", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0663_02", "gt_frame_id": "690", "anchor_frame_id": "2093", "shot": 6, "contexts_frame_id": ["2093", "2118", "2480", "2082", "2734", "758"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001522", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0665_01", "gt_frame_id": "971", "anchor_frame_id": "6", "shot": 6, "contexts_frame_id": ["6", "930", "172", "212", "280", "344"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001525", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0685_01", "gt_frame_id": "540", "anchor_frame_id": "487", "shot": 6, "contexts_frame_id": ["487", "390", "198", "588", "656", "842"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the shelf, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the shelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001528", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0695_01", "gt_frame_id": "2025", "anchor_frame_id": "1141", "shot": 6, "contexts_frame_id": ["1141", "1474", "1934", "594", "612", "1200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001531", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0695_03", "gt_frame_id": "2085", "anchor_frame_id": "1217", "shot": 6, "contexts_frame_id": ["1217", "652", "1036", "348", "2238", "2442"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001534", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0696_00", "gt_frame_id": "413", "anchor_frame_id": "1105", "shot": 6, "contexts_frame_id": ["1105", "520", "454", "356", "1048", "194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001537", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0697_02", "gt_frame_id": "2814", "anchor_frame_id": "1384", "shot": 6, "contexts_frame_id": ["1384", "1214", "1654", "180", "2444", "2420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the nightstand, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the nightstand, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001540", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0700_01", "gt_frame_id": "694", "anchor_frame_id": "738", "shot": 6, "contexts_frame_id": ["738", "114", "174", "228", "282", "3252"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the file cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the file cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001543", "dataset": "scannet", "task_type": "anchor_guided_relative_repositioning", "scene_id": "scene0704_01", "gt_frame_id": "1224", "anchor_frame_id": "164", "shot": 6, "contexts_frame_id": ["164", "1786", "1266", "486", "574", "82"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the printer, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the printer, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001546", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0568_00", "gt_frame_id": "1116", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["566", "620", "678", "866", "916", "1146"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table between two chairs.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001549", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0568_02", "gt_frame_id": "885", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["920", "838", "800", "668", "634", "584"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe curtain above the radiator, and to the left of the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001552", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0412_00", "gt_frame_id": "1091", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1138", "390", "414", "1312", "1454", "1654"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe recycling bin to the right of three trash cans.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001555", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0217_00", "gt_frame_id": "375", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1078", "1134", "336", "1180", "1208", "422"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair between the window and the backpack.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001558", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0426_00", "gt_frame_id": "1412", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1164", "1310", "80", "100", "108", "144"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe desk under the keyboard piano, and under the monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001561", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0050_01", "gt_frame_id": "2780", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2820", "1192", "1174", "2350", "1508", "1644"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe laptop on the desk, and to the right of the lamp.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001564", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0169_01", "gt_frame_id": "624", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["848", "782", "702", "690", "582", "1492"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet to the left of two chairs.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 3.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001567", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0221_01", "gt_frame_id": "243", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["622", "486", "318", "276", "202", "450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe lamp on the nightstand, and to the right of the pillow.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001570", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0164_00", "gt_frame_id": "195", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1592", "528", "428", "402", "226", "1278"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe stove below the range hood, and to the right of the box.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001573", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0599_01", "gt_frame_id": "17", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2068", "770", "2188", "66", "44", "1254"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe door to the right of the doorframe, and to the right of the trash can.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001576", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_00", "gt_frame_id": "2147", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2876", "3012", "3096", "1932", "2060", "3352"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe guitar to the right of the coffee table, and to the right of the plant.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001579", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_00", "gt_frame_id": "3346", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3184", "2280", "2202", "4", "334", "486"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe plant on the coffee table, and to the left of the guitar.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001582", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_00", "gt_frame_id": "3305", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3024", "3102", "3136", "3186", "3254", "2232"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe coffee table under the plant, and to the left of the guitar.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001585", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_02", "gt_frame_id": "1819", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["912", "462", "666", "2032", "1872", "1778"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the right of the guitar, and to the right of the keyboard piano.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001588", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_02", "gt_frame_id": "326", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2688", "2618", "2312", "724", "378", "282"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe tv on the tv stand, and under the plant.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001591", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_02", "gt_frame_id": "1970", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["322", "264", "208", "1840", "2844", "2798"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe keyboard piano to the right of the plant, and to the right of the coffee table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001594", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0608_02", "gt_frame_id": "78", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2740", "1700", "1830", "2826", "1992", "240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe guitar to the right of the coffee table, and to the right of the plant.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001597", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0357_00", "gt_frame_id": "96", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["186", "138", "700", "890", "732", "756"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe pipe between the shelf and the curtain.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001600", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0357_01", "gt_frame_id": "182", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["906", "882", "820", "702", "84", "608"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bag between the shelf and the curtain, and to the left of the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001603", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0357_01", "gt_frame_id": "189", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["702", "152", "832", "882", "270", "290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe pipe between the shelf and the curtain, and to the left of the window.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001606", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0695_02", "gt_frame_id": "747", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["380", "1116", "10", "528", "548", "576"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe recycling bin between the trash can and two boxes.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001609", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0695_03", "gt_frame_id": "650", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["216", "1172", "816", "596", "772", "2442"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe box to the right of the bookshelf, and behind the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001612", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0629_01", "gt_frame_id": "458", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2052", "604", "672", "1172", "988", "1286"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe door between two pictures.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001615", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0629_01", "gt_frame_id": "551", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2064", "606", "1214", "794", "970", "1270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe doorframe between two pictures.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001618", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0629_01", "gt_frame_id": "1837", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["598", "500", "268", "1962", "1932", "1716"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe picture to the right of the plant, and to the right of the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001621", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0629_02", "gt_frame_id": "1487", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["246", "204", "88", "56", "1398", "1386"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe plant on the table, and to the left of the picture.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001624", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0633_01", "gt_frame_id": "1380", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1116", "326", "914", "444", "796", "630"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the left of the curtain, to the left of the table, and to the left of the window.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001627", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0633_01", "gt_frame_id": "891", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["248", "206", "166", "1080", "972", "432"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table under the window, and to the right of the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001630", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0696_01", "gt_frame_id": "961", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1110", "1086", "1030", "1014", "902", "288"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe radiator on the couch, and under the curtain.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 3.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001633", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0696_01", "gt_frame_id": "203", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1354", "1412", "144", "832", "892", "288"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet to the left of the box, and to the left of the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 3.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001636", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0696_01", "gt_frame_id": "119", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1158", "1208", "1286", "674", "8", "782"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe shelf between the box and the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001639", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0645_00", "gt_frame_id": "2696", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3036", "3660", "3714", "3772", "2838", "2612"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the tv, and to the right of the suitcase.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001642", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0645_01", "gt_frame_id": "1314", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["606", "1590", "1474", "1218", "1848", "2972"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe pillow on the coffee table, and in front of the couch.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001645", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0645_02", "gt_frame_id": "1751", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2040", "1940", "1804", "1466", "1346", "1710"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe couch behind the coffee table, and behind the pillow.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001648", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0645_02", "gt_frame_id": "2788", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2480", "2542", "2574", "2668", "2744", "170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe nightstand under the lamp, and to the left of the pillow.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001651", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0353_00", "gt_frame_id": "1509", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1442", "1586", "220", "256", "1722", "518"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bookshelf between the trash can and the whiteboard.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001654", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0353_00", "gt_frame_id": "94", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1286", "1250", "2796", "2874", "2916", "1536"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe whiteboard to the right of the bookshelf, and behind the pillow.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001657", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0353_01", "gt_frame_id": "1452", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["432", "378", "352", "138", "82", "1412"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bookshelf to the left of the board, and behind the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001660", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0353_02", "gt_frame_id": "1272", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1476", "1434", "1380", "2100", "1196", "16"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bookshelf between the trash can and the whiteboard.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001663", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0593_00", "gt_frame_id": "2115", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["222", "292", "2158", "1942", "2010", "450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe trash can between the paper cutter and the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001666", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0246_00", "gt_frame_id": "364", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1860", "12", "334", "892", "956", "992"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe mirror on the table, and under the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001669", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0653_01", "gt_frame_id": "491", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3582", "540", "3356", "376", "4782", "202"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe desk between the whiteboard and the office chair, and under the box.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001672", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0653_01", "gt_frame_id": "536", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["296", "3294", "3452", "482", "3686", "766"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe whiteboard to the left of the box, and to the left of the desk.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001675", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0653_01", "gt_frame_id": "1855", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2494", "2214", "2166", "1508", "1550", "4322"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe box behind the file cabinet, and to the right of the office chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001678", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0356_00", "gt_frame_id": "653", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["248", "204", "864", "836", "814", "54"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe recycling bin on the desk, on the chair, and to the right of the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001681", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0356_00", "gt_frame_id": "37", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["268", "948", "178", "104", "1342", "1314"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe trash bin on the desk, and to the right of the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001684", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0356_01", "gt_frame_id": "107", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["346", "894", "812", "760", "76", "1282"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe trash can to the right of the recycling bin, and to the right of the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001687", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0378_02", "gt_frame_id": "1384", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1026", "1196", "1244", "1358", "1472", "1656"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe door to the left of the bookshelf, and to the left of the book.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001690", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0553_02", "gt_frame_id": "1517", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["910", "1032", "1346", "1412", "116", "162"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe toilet to the left of the toilet paper dispenser, and to the left of the bar.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001693", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0663_00", "gt_frame_id": "616", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2916", "482", "542", "1752", "644", "668"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe shelf above two monitors.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001696", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0663_01", "gt_frame_id": "1925", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["414", "154", "3112", "2454", "2062", "1820"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe jacket on the suitcase, and to the left of the backpack.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001699", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0207_02", "gt_frame_id": "964", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1000", "352", "2234", "2420", "94", "2508"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe clothes between two backpacks, and to the right of the stool.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001702", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0550_00", "gt_frame_id": "1731", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1324", "80", "1532", "162", "1642", "1872"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe paper towel dispenser under two toilet papers.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001705", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0648_00", "gt_frame_id": "2419", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1646", "1318", "270", "1084", "1058", "3544"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe backpack to the left of the fan, and to the left of the shoe.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001708", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0435_00", "gt_frame_id": "2485", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1438", "3012", "2816", "2784", "2742", "2542"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe toilet between the trash can and the shower curtain.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001711", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0435_03", "gt_frame_id": "78", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1798", "206", "1046", "1098", "1246", "1628"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe office chair to the right of the desk, and to the right of the trash can.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001714", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0697_00", "gt_frame_id": "182", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1502", "376", "340", "1344", "1204", "1264"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe storage bin to the right of the nightstand, and to the right of the tissue box.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001717", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0697_03", "gt_frame_id": "1135", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["638", "2562", "2644", "2680", "1186", "1744"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe board on the box, below the picture, and to the left of the dresser.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001720", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0697_03", "gt_frame_id": "512", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1734", "1516", "1600", "456", "2372", "576"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe lamp on the nightstand, and to the left of the storage bin.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001723", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0474_05", "gt_frame_id": "1675", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2830", "1632", "3002", "3032", "1974", "2012"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe file cabinet between the computer tower and the trash can.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001726", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0146_02", "gt_frame_id": "330", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1124", "1294", "164", "1420", "394", "440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe rail between the toilet and the sink.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001729", "dataset": "scannet", "task_type": "obj_loc", "scene_id": "scene0100_02", "gt_frame_id": "698", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["468", "366", "502", "584", "832", "2"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe mat under the toilet, and to the right of the bathtub.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001732", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0025_01", "gt_frame_id": "1242", "anchor_frame_id": "1312", "shot": 6, "contexts_frame_id": ["1312", "564", "720", "762", "330", "30"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the pillow.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001735", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0030_00", "gt_frame_id": "579", "anchor_frame_id": "703", "shot": 6, "contexts_frame_id": ["703", "620", "1970", "1882", "770", "904"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bag.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001738", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0030_01", "gt_frame_id": "1601", "anchor_frame_id": "652", "shot": 6, "contexts_frame_id": ["652", "672", "548", "582", "50", "1548"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the armchair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001741", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0030_02", "gt_frame_id": "483", "anchor_frame_id": "356", "shot": 6, "contexts_frame_id": ["356", "400", "456", "508", "558", "702"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tray near the box.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001744", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0030_02", "gt_frame_id": "1178", "anchor_frame_id": "287", "shot": 6, "contexts_frame_id": ["287", "334", "264", "140", "68", "26"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the armchair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001747", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0046_01", "gt_frame_id": "0", "anchor_frame_id": "157", "shot": 6, "contexts_frame_id": ["157", "126", "2268", "78", "618", "496"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the nightstand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001750", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0046_01", "gt_frame_id": "705", "anchor_frame_id": "242", "shot": 6, "contexts_frame_id": ["242", "146", "1928", "1872", "478", "2186"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the clothes.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001753", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0046_01", "gt_frame_id": "2386", "anchor_frame_id": "499", "shot": 6, "contexts_frame_id": ["499", "684", "418", "378", "218", "264"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the laptop.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001756", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0046_02", "gt_frame_id": "620", "anchor_frame_id": "700", "shot": 6, "contexts_frame_id": ["700", "2476", "916", "2396", "1956", "2228"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the clothes.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001759", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0046_02", "gt_frame_id": "849", "anchor_frame_id": "331", "shot": 6, "contexts_frame_id": ["331", "232", "2058", "2026", "1944", "652"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the clothes.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001762", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0050_02", "gt_frame_id": "817", "anchor_frame_id": "1888", "shot": 6, "contexts_frame_id": ["1888", "3038", "3492", "1558", "1696", "1734"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tissue box.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001765", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0084_00", "gt_frame_id": "617", "anchor_frame_id": "557", "shot": 6, "contexts_frame_id": ["557", "580", "116", "1914", "1728", "1662"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sign.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001768", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0084_01", "gt_frame_id": "821", "anchor_frame_id": "1124", "shot": 6, "contexts_frame_id": ["1124", "1190", "550", "598", "710", "890"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the shower head.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001771", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0100_00", "gt_frame_id": "854", "anchor_frame_id": "520", "shot": 6, "contexts_frame_id": ["520", "486", "682", "750", "798", "230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001774", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0100_02", "gt_frame_id": "750", "anchor_frame_id": "203", "shot": 6, "contexts_frame_id": ["203", "144", "88", "810", "692", "564"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001777", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0146_00", "gt_frame_id": "219", "anchor_frame_id": "402", "shot": 6, "contexts_frame_id": ["402", "524", "690", "786", "314", "1386"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001780", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0146_00", "gt_frame_id": "386", "anchor_frame_id": "201", "shot": 6, "contexts_frame_id": ["201", "270", "342", "512", "1370", "1334"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001783", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0146_02", "gt_frame_id": "373", "anchor_frame_id": "169", "shot": 6, "contexts_frame_id": ["169", "1342", "1262", "1064", "486", "332"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001786", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0164_02", "gt_frame_id": "119", "anchor_frame_id": "268", "shot": 6, "contexts_frame_id": ["268", "228", "1170", "1100", "20", "1048"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the range hood.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001789", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0207_02", "gt_frame_id": "319", "anchor_frame_id": "437", "shot": 6, "contexts_frame_id": ["437", "90", "2506", "2234", "964", "928"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the book.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001792", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0222_01", "gt_frame_id": "3778", "anchor_frame_id": "3306", "shot": 6, "contexts_frame_id": ["3306", "3714", "1968", "3878", "2336", "2714"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the fan.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001795", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0222_01", "gt_frame_id": "2014", "anchor_frame_id": "2099", "shot": 6, "contexts_frame_id": ["2099", "850", "1012", "3770", "2970", "3046"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the radiator.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001798", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0231_02", "gt_frame_id": "185", "anchor_frame_id": "2848", "shot": 6, "contexts_frame_id": ["2848", "2364", "2316", "644", "2210", "678"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the fireplace.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001801", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0231_02", "gt_frame_id": "2404", "anchor_frame_id": "744", "shot": 6, "contexts_frame_id": ["744", "324", "2312", "40", "554", "836"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the backpack near the trash can.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001804", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0277_01", "gt_frame_id": "190", "anchor_frame_id": "288", "shot": 6, "contexts_frame_id": ["288", "226", "144", "806", "410", "456"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the radiator.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001807", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0307_02", "gt_frame_id": "2027", "anchor_frame_id": "246", "shot": 6, "contexts_frame_id": ["246", "640", "2054", "164", "546", "0"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the chair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001810", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0307_02", "gt_frame_id": "1814", "anchor_frame_id": "930", "shot": 6, "contexts_frame_id": ["930", "1002", "246", "282", "1782", "1854"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the broom.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001813", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0328_00", "gt_frame_id": "75", "anchor_frame_id": "199", "shot": 6, "contexts_frame_id": ["199", "1104", "1352", "1370", "1396", "298"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the microwave.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001816", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0351_01", "gt_frame_id": "800", "anchor_frame_id": "709", "shot": 6, "contexts_frame_id": ["709", "722", "836", "468", "438", "342"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the radiator.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001819", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0353_00", "gt_frame_id": "2276", "anchor_frame_id": "2668", "shot": 6, "contexts_frame_id": ["2668", "888", "2400", "792", "702", "2082"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the laptop.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001822", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0356_02", "gt_frame_id": "553", "anchor_frame_id": "125", "shot": 6, "contexts_frame_id": ["125", "114", "82", "710", "180", "224"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the recycling bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001825", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0378_01", "gt_frame_id": "1097", "anchor_frame_id": "1650", "shot": 6, "contexts_frame_id": ["1650", "1558", "1746", "1244", "1184", "1038"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the paper near the box.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001828", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0382_01", "gt_frame_id": "353", "anchor_frame_id": "318", "shot": 6, "contexts_frame_id": ["318", "180", "714", "648", "66", "114"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the dresser.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001831", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0430_00", "gt_frame_id": "2027", "anchor_frame_id": "55", "shot": 6, "contexts_frame_id": ["55", "1764", "1344", "1202", "2106", "1040"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the backpack.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001834", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0430_00", "gt_frame_id": "2106", "anchor_frame_id": "2215", "shot": 6, "contexts_frame_id": ["2215", "1332", "1218", "1992", "1964", "1908"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the backpack.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001837", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_00", "gt_frame_id": "737", "anchor_frame_id": "956", "shot": 6, "contexts_frame_id": ["956", "1170", "1402", "778", "660", "614"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sofa chair near the clothes.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001840", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_00", "gt_frame_id": "2495", "anchor_frame_id": "2641", "shot": 6, "contexts_frame_id": ["2641", "2542", "2742", "2784", "2816", "3012"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001843", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_00", "gt_frame_id": "187", "anchor_frame_id": "1498", "shot": 6, "contexts_frame_id": ["1498", "1438", "1582", "1626", "1776", "70"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the office chair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001846", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_01", "gt_frame_id": "1397", "anchor_frame_id": "200", "shot": 6, "contexts_frame_id": ["200", "1492", "3188", "1626", "3144", "1290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv stand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001849", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_01", "gt_frame_id": "3167", "anchor_frame_id": "73", "shot": 6, "contexts_frame_id": ["73", "226", "1494", "1476", "1366", "1290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv stand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001852", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_01", "gt_frame_id": "73", "anchor_frame_id": "3160", "shot": 6, "contexts_frame_id": ["3160", "1686", "1764", "1488", "1534", "186"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the chair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001855", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0435_03", "gt_frame_id": "782", "anchor_frame_id": "1489", "shot": 6, "contexts_frame_id": ["1489", "1564", "1362", "736", "588", "868"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the nightstand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001858", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0552_01", "gt_frame_id": "247", "anchor_frame_id": "138", "shot": 6, "contexts_frame_id": ["138", "336", "222", "306", "294", "278"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toaster oven.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001861", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0552_01", "gt_frame_id": "407", "anchor_frame_id": "354", "shot": 6, "contexts_frame_id": ["354", "112", "106", "96", "44", "32"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bucket.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001864", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0553_00", "gt_frame_id": "1112", "anchor_frame_id": "561", "shot": 6, "contexts_frame_id": ["561", "808", "1020", "1156", "1458", "418"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001867", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0553_01", "gt_frame_id": "966", "anchor_frame_id": "707", "shot": 6, "contexts_frame_id": ["707", "788", "892", "1156", "1054", "1296"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001870", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0568_00", "gt_frame_id": "914", "anchor_frame_id": "1229", "shot": 6, "contexts_frame_id": ["1229", "1284", "1164", "1006", "954", "876"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the radiator.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001873", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0578_01", "gt_frame_id": "869", "anchor_frame_id": "322", "shot": 6, "contexts_frame_id": ["322", "270", "64", "574", "516", "450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the backpack.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001876", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0583_01", "gt_frame_id": "918", "anchor_frame_id": "1325", "shot": 6, "contexts_frame_id": ["1325", "1518", "246", "228", "294", "954"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the stool.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001879", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0593_00", "gt_frame_id": "1749", "anchor_frame_id": "1893", "shot": 6, "contexts_frame_id": ["1893", "1222", "1178", "1028", "1682", "1422"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bulletin board.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001882", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0606_00", "gt_frame_id": "791", "anchor_frame_id": "1013", "shot": 6, "contexts_frame_id": ["1013", "852", "758", "726", "1128", "1336"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the pillow.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001885", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0606_01", "gt_frame_id": "2822", "anchor_frame_id": "1790", "shot": 6, "contexts_frame_id": ["1790", "2616", "2224", "2112", "338", "494"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001888", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0606_02", "gt_frame_id": "2480", "anchor_frame_id": "1817", "shot": 6, "contexts_frame_id": ["1817", "1876", "2122", "2516", "260", "488"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bag near the storage bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001891", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0608_00", "gt_frame_id": "419", "anchor_frame_id": "936", "shot": 6, "contexts_frame_id": ["936", "826", "636", "516", "338", "332"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001894", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0608_01", "gt_frame_id": "1907", "anchor_frame_id": "1769", "shot": 6, "contexts_frame_id": ["1769", "1818", "2094", "2744", "112", "166"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the guitar.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001897", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0608_02", "gt_frame_id": "318", "anchor_frame_id": "926", "shot": 6, "contexts_frame_id": ["926", "1104", "1170", "750", "368", "278"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001900", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0629_00", "gt_frame_id": "2739", "anchor_frame_id": "545", "shot": 6, "contexts_frame_id": ["545", "620", "288", "1938", "102", "2692"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the plant.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001903", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0629_00", "gt_frame_id": "2633", "anchor_frame_id": "2545", "shot": 6, "contexts_frame_id": ["2545", "2350", "2592", "1382", "2670", "1792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the chair near the plant.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001906", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0629_01", "gt_frame_id": "1790", "anchor_frame_id": "317", "shot": 6, "contexts_frame_id": ["317", "246", "180", "1832", "2610", "2586"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the plant.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001909", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0629_02", "gt_frame_id": "1487", "anchor_frame_id": "291", "shot": 6, "contexts_frame_id": ["291", "192", "88", "56", "1398", "1386"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the plant near the picture.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001912", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0633_01", "gt_frame_id": "469", "anchor_frame_id": "586", "shot": 6, "contexts_frame_id": ["586", "530", "802", "1344", "948", "346"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the lamp.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001915", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0633_01", "gt_frame_id": "947", "anchor_frame_id": "1253", "shot": 6, "contexts_frame_id": ["1253", "1266", "1088", "408", "486", "240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the window.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001918", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0643_00", "gt_frame_id": "1222", "anchor_frame_id": "1400", "shot": 6, "contexts_frame_id": ["1400", "1822", "1874", "1376", "1196", "1162"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the telephone.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001921", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_00", "gt_frame_id": "1684", "anchor_frame_id": "2669", "shot": 6, "contexts_frame_id": ["2669", "2804", "3310", "2530", "1726", "1426"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the chair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001924", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_01", "gt_frame_id": "2118", "anchor_frame_id": "2777", "shot": 6, "contexts_frame_id": ["2777", "3664", "3202", "1936", "2024", "2318"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trash can near the recycling bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001927", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_01", "gt_frame_id": "3587", "anchor_frame_id": "3426", "shot": 6, "contexts_frame_id": ["3426", "3330", "3040", "5170", "168", "276"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the nightstand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001930", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_01", "gt_frame_id": "1806", "anchor_frame_id": "1355", "shot": 6, "contexts_frame_id": ["1355", "1452", "1512", "1764", "2028", "3252"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the chair.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001933", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_02", "gt_frame_id": "2791", "anchor_frame_id": "329", "shot": 6, "contexts_frame_id": ["329", "228", "198", "162", "2744", "2658"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the nightstand.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001936", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0645_02", "gt_frame_id": "1024", "anchor_frame_id": "1434", "shot": 6, "contexts_frame_id": ["1434", "1928", "2056", "1308", "1282", "876"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trash can near the recycling bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001939", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0651_02", "gt_frame_id": "238", "anchor_frame_id": "150", "shot": 6, "contexts_frame_id": ["150", "738", "186", "280", "308", "42"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001942", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0663_01", "gt_frame_id": "2695", "anchor_frame_id": "2474", "shot": 6, "contexts_frame_id": ["2474", "2042", "2352", "1936", "1736", "1654"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the jacket.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001945", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0693_00", "gt_frame_id": "357", "anchor_frame_id": "718", "shot": 6, "contexts_frame_id": ["718", "564", "466", "214", "1574", "1470"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the sink.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001948", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0695_00", "gt_frame_id": "751", "anchor_frame_id": "793", "shot": 6, "contexts_frame_id": ["793", "828", "1646", "1776", "716", "948"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the box near the backpack.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001951", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0695_02", "gt_frame_id": "997", "anchor_frame_id": "444", "shot": 6, "contexts_frame_id": ["444", "162", "402", "1274", "266", "1106"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the laptop.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001954", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0695_02", "gt_frame_id": "1243", "anchor_frame_id": "1205", "shot": 6, "contexts_frame_id": ["1205", "1006", "1274", "1498", "1916", "1896"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bag.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001957", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0695_03", "gt_frame_id": "2084", "anchor_frame_id": "1298", "shot": 6, "contexts_frame_id": ["1298", "854", "260", "652", "1204", "2440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bookshelf.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001960", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0697_03", "gt_frame_id": "150", "anchor_frame_id": "1211", "shot": 6, "contexts_frame_id": ["1211", "1198", "1752", "2660", "2604", "668"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the picture.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001963", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0697_03", "gt_frame_id": "2414", "anchor_frame_id": "2320", "shot": 6, "contexts_frame_id": ["2320", "2382", "576", "636", "2548", "40"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the closet door.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001966", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0697_03", "gt_frame_id": "1143", "anchor_frame_id": "958", "shot": 6, "contexts_frame_id": ["958", "44", "2702", "1186", "640", "2584"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the board near the picture.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001969", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0697_03", "gt_frame_id": "512", "anchor_frame_id": "597", "shot": 6, "contexts_frame_id": ["597", "636", "576", "2372", "456", "1600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the lamp near the tissue box.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001972", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0700_01", "gt_frame_id": "47", "anchor_frame_id": "425", "shot": 6, "contexts_frame_id": ["425", "1212", "606", "1084", "2466", "2278"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the keyboard.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001975", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0700_01", "gt_frame_id": "109", "anchor_frame_id": "1302", "shot": 6, "contexts_frame_id": ["1302", "1212", "1044", "3966", "3864", "714"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the paper.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001978", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0700_02", "gt_frame_id": "578", "anchor_frame_id": "337", "shot": 6, "contexts_frame_id": ["337", "1818", "1956", "3354", "3434", "3544"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the file cabinet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001981", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0700_02", "gt_frame_id": "3127", "anchor_frame_id": "3068", "shot": 6, "contexts_frame_id": ["3068", "2634", "1118", "1298", "826", "154"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the telephone.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001984", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0700_02", "gt_frame_id": "1056", "anchor_frame_id": "3474", "shot": 6, "contexts_frame_id": ["3474", "3396", "3370", "3652", "1266", "1300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the keyboard near the monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001987", "dataset": "scannet", "task_type": "loc-re-facing", "scene_id": "scene0704_01", "gt_frame_id": "164", "anchor_frame_id": "100", "shot": 6, "contexts_frame_id": ["100", "84", "1322", "210", "128", "1226"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the printer.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001990", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0025_01", "gt_frame_id": "428", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["476", "1358", "1316", "552", "1242", "1226"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the fan, the file cabinet is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001993", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0025_01", "gt_frame_id": "520", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1170", "122", "1226", "1242", "550", "446"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the file cabinet, I am facing the fan, and I can see the door on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001996", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_00", "gt_frame_id": "379", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1038", "2366", "2420", "108", "2444", "348"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the tv, and the clothes remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00001999", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_00", "gt_frame_id": "2193", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["702", "2338", "2298", "2234", "1986", "1872"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the doorframe, I am facing the towel, and I can see the mat near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002002", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_00", "gt_frame_id": "1037", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["152", "2436", "24", "2380", "2344", "2304"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the chair, I face the table.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002005", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_00", "gt_frame_id": "2472", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["114", "2422", "2352", "962", "260", "360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the clothes to my left, I am facing the nightstand.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002008", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_00", "gt_frame_id": "776", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["716", "672", "2402", "228", "348", "310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the chair, I am facing the pillow.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002011", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_01", "gt_frame_id": "3", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2124", "2256", "126", "192", "1908", "1886"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the clothes to my left, I am facing the nightstand.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002014", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_01", "gt_frame_id": "1775", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1926", "1864", "1836", "1818", "1054", "1242"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the doorframe, I am facing the towel.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002017", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_01", "gt_frame_id": "1947", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["620", "492", "684", "1900", "184", "240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the clothes, I am facing the table.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002020", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_02", "gt_frame_id": "1832", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["924", "962", "1110", "1154", "1194", "1874"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the doorframe, facing the towel, and I can see the sink on the right side of the view, and the mat near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002023", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0046_02", "gt_frame_id": "618", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2114", "2214", "1974", "2396", "870", "912"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the chair, I face the clothes, and I can see the bag on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002026", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0050_00", "gt_frame_id": "4265", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3994", "298", "356", "2038", "2094", "3582"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the cup, the armchair is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002029", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0050_00", "gt_frame_id": "664", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2904", "2728", "4608", "828", "1200", "1298"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the stool, with the ottoman to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002032", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0050_01", "gt_frame_id": "2826", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3546", "3476", "1196", "3308", "2786", "3206"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the chair, facing the cup.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002035", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0050_02", "gt_frame_id": "2431", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3102", "358", "348", "2538", "296", "3830"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the chair, I face the printer.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002038", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0050_02", "gt_frame_id": "2706", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2930", "300", "292", "3892", "3096", "3058"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the sofa chair, I am facing the shelf.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002041", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0100_01", "gt_frame_id": "180", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["554", "504", "418", "998", "118", "246"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the toilet to my right, I face the soap dish.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002044", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0100_02", "gt_frame_id": "986", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["116", "26", "704", "592", "660", "952"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the stool, I am facing the soap dish.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002047", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0100_02", "gt_frame_id": "821", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["136", "86", "780", "716", "580", "624"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the toilet, I am facing the toilet paper, and I can see the bathtub on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002050", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0144_00", "gt_frame_id": "69", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["132", "666", "690", "1356", "1122", "1194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the armchair behind me, I am facing the lamp.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002053", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0207_01", "gt_frame_id": "314", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["204", "1770", "1794", "1838", "428", "372"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the laptop to my left, I face the suitcase.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002056", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0207_02", "gt_frame_id": "1690", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1614", "1568", "1488", "1110", "1912", "1862"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the doorframe, I face the toilet, and I can see the bathroom vanity on the left side of the view, and the bathtub on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002059", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0207_02", "gt_frame_id": "2587", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["732", "710", "0", "462", "2148", "2238"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the laundry basket, the chair is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002062", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0207_02", "gt_frame_id": "2478", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2508", "442", "2286", "2432", "158", "204"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the chair to my left, I am facing the printer.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002065", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0222_01", "gt_frame_id": "2216", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2730", "2400", "290", "256", "746", "2100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the bottle, with the lamp to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002068", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0231_00", "gt_frame_id": "3312", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1830", "4200", "1498", "4434", "3194", "3078"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the stool to my right, I am facing the poster.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002071", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0277_01", "gt_frame_id": "195", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["410", "282", "874", "842", "126", "736"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the dresser, I am facing the radiator, and I can see the desk on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002074", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0329_00", "gt_frame_id": "697", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["758", "726", "670", "650", "20", "1148"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the coffee table, I am facing the lamp.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002077", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0353_00", "gt_frame_id": "1848", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["912", "748", "694", "672", "610", "568"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am at a position where the mini fridge is behind me and the plate is in front of me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002080", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0353_00", "gt_frame_id": "646", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["912", "894", "744", "686", "1980", "562"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the plate, the mini fridge is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002083", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0353_01", "gt_frame_id": "666", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["986", "948", "800", "698", "1830", "1806"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the mini fridge to my right, I face the plate.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002086", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0357_00", "gt_frame_id": "67", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["890", "116", "194", "128", "662", "326"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the window, the table is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002089", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0357_00", "gt_frame_id": "697", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["638", "672", "120", "96", "882", "732"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the table, I face the backpack.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002092", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0357_00", "gt_frame_id": "24", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["646", "678", "116", "716", "732", "882"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the table, facing the bag.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002095", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0357_01", "gt_frame_id": "182", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["84", "702", "758", "782", "832", "212"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the table to my left, I face the bag.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002098", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0406_00", "gt_frame_id": "675", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["616", "330", "48", "740", "796", "860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the toilet paper, and the counter is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002101", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0406_01", "gt_frame_id": "1115", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["974", "1026", "1084", "14", "1170", "64"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the toilet, I face the toilet paper.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002104", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0426_00", "gt_frame_id": "13", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1212", "1280", "1350", "1468", "216", "132"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the keyboard piano, with the bed to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002107", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0435_01", "gt_frame_id": "554", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1532", "190", "334", "398", "508", "612"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nThe chair is behind me as I face the clothes.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002110", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0435_01", "gt_frame_id": "540", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["188", "356", "432", "496", "614", "734"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the chair to my left, I face the clothes.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002113", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0435_01", "gt_frame_id": "703", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["324", "372", "508", "578", "942", "744"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the mini fridge, I am facing the backpack.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002116", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0553_01", "gt_frame_id": "102", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["814", "904", "1038", "1262", "1324", "20"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the toilet paper, the sink is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002119", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0578_01", "gt_frame_id": "582", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["242", "382", "892", "540", "4", "56"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am oriented toward the backpack, with the trash can behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002122", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0583_01", "gt_frame_id": "887", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["466", "1524", "202", "228", "258", "834"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the stool, with the shower curtain to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002125", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0583_01", "gt_frame_id": "836", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1564", "1386", "68", "80", "316", "888"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the toilet, I am facing the bottle, and I can see the stool on the right side of the view, and the trash can near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002128", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0583_02", "gt_frame_id": "1423", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1100", "924", "244", "788", "378", "566"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the doorframe, I am facing the picture, and I can see the window on the left side of the view, and the mirror on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002131", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0583_02", "gt_frame_id": "266", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1034", "910", "96", "1546", "344", "778"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the doorframe, facing the bottle.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002134", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0583_02", "gt_frame_id": "853", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["402", "1454", "262", "58", "82", "880"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the toilet, I face the bottle, and I can see the toilet paper near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002137", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0591_01", "gt_frame_id": "640", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["690", "516", "488", "434", "376", "1416"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the telephone, the backpack is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002140", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0606_02", "gt_frame_id": "1965", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["600", "540", "300", "84", "2520", "2116"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the tray, I am facing the bag.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002143", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0608_01", "gt_frame_id": "2064", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["678", "2716", "2186", "1864", "146", "14"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the couch behind me, I am facing the guitar.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002146", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0633_00", "gt_frame_id": "80", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1544", "132", "106", "528", "826", "634"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the lamp, and the wardrobe is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002149", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0633_00", "gt_frame_id": "102", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["300", "276", "144", "468", "0", "618"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the window, the ball is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002152", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0645_01", "gt_frame_id": "2893", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1252", "3276", "3264", "1764", "2024", "2174"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the recycling bin, and the coffee table is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002155", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0645_01", "gt_frame_id": "2436", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2592", "2526", "1990", "2010", "2060", "2318"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the recycling bin, the chair is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002158", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0645_02", "gt_frame_id": "1985", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2294", "1944", "1830", "1768", "1728", "1688"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the coffee table to my left, I face the end table.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002161", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0647_01", "gt_frame_id": "815", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["648", "702", "724", "754", "786", "844"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the chair to my left, I face the tv.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002164", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0664_00", "gt_frame_id": "1213", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1128", "1170", "1254", "278", "478", "554"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the toilet, I face the toilet paper, and I can see the bathroom vanity on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002167", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0664_01", "gt_frame_id": "1247", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1166", "1282", "1292", "58", "396", "570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the toilet to my left, I face the shelf.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002170", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0664_01", "gt_frame_id": "1280", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["636", "570", "396", "8", "78", "1236"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the toilet, facing the toilet paper, and I can see the bathroom counter on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002173", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0693_00", "gt_frame_id": "939", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1572", "210", "544", "812", "900", "1130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nThe bathroom vanity is behind me as I face the bottle.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002176", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0693_02", "gt_frame_id": "167", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1224", "1350", "1388", "52", "1476", "418"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the mat, I am facing the soap dish, and I can see the sink near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002179", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0695_01", "gt_frame_id": "1757", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["604", "940", "1142", "976", "1460", "1806"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the mat behind me, I am facing the laptop.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002182", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0695_02", "gt_frame_id": "1628", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1012", "1226", "1682", "1676", "1914", "1498"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the paper bag, I face the bag, and I can see the laptop on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002185", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0695_02", "gt_frame_id": "1963", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2036", "1118", "388", "1292", "1692", "1922"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the paper bag to my left, I face the office chair.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002188", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0695_03", "gt_frame_id": "2064", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1224", "2024", "828", "590", "772", "2442"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the mat, I am facing the bag, and I can see the book on the left side of the view, and the case of water bottles near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002191", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0697_00", "gt_frame_id": "68", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2804", "938", "1182", "1196", "1322", "246"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the tissue box, and the bed is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002194", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0700_00", "gt_frame_id": "2449", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1958", "2306", "2340", "2368", "2944", "3766"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the telephone, with the armchair to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002197", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0700_00", "gt_frame_id": "2910", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2442", "3794", "3738", "2602", "158", "3558"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the keyboard, with the armchair behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002200", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0700_00", "gt_frame_id": "2945", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2450", "3694", "2902", "3618", "2308", "1960"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the armchair behind me, I am facing the telephone.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002203", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0700_02", "gt_frame_id": "88", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3286", "3228", "3652", "1050", "3112", "2622"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the file cabinet to my left, I face the keyboard.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002206", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0702_00", "gt_frame_id": "365", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["648", "612", "574", "400", "262", "280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the trash can, with the door to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002209", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0702_02", "gt_frame_id": "206", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["540", "426", "406", "2", "70", "120"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the toilet, facing the trash can.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002212", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0702_02", "gt_frame_id": "86", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["540", "426", "22", "396", "200", "170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the door, I am facing the telephone.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002215", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0702_02", "gt_frame_id": "170", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["426", "406", "206", "52", "88", "100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the telephone, the toilet is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002218", "dataset": "scannet", "task_type": "view_localization", "scene_id": "scene0704_00", "gt_frame_id": "1254", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["1294", "1856", "580", "1210", "522", "486"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the copier to my left, I face the water cooler.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002221", "dataset": "scannetpp", "task_type": "composition", "scene_id": "09c1414f1b", "gt_frame_id": "frame_008400", "anchor_frame_id": "frame_007890", "shot": 6, "contexts_frame_id": ["frame_007890", "frame_007790", "frame_007550", "frame_008030", "frame_006250", "frame_004240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002224", "dataset": "scannetpp", "task_type": "composition", "scene_id": "09c1414f1b", "gt_frame_id": "frame_007560", "anchor_frame_id": "frame_008400", "shot": 6, "contexts_frame_id": ["frame_008400", "frame_008310", "frame_008410", "frame_008510", "frame_008530", "frame_007310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002227", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "09c1414f1b", "gt_frame_id": "frame_008170", "anchor_frame_id": "frame_008300", "shot": 6, "contexts_frame_id": ["frame_008300", "frame_007870", "frame_007770", "frame_007510", "frame_005840", "frame_004940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002230", "dataset": "scannetpp", "task_type": "composition", "scene_id": "09c1414f1b", "gt_frame_id": "frame_006610", "anchor_frame_id": "frame_005430", "shot": 6, "contexts_frame_id": ["frame_005430", "frame_005180", "frame_005160", "frame_004940", "frame_007190", "frame_007310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 2.1 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002233", "dataset": "scannetpp", "task_type": "translation", "scene_id": "0d2ee665be", "gt_frame_id": "frame_003780", "anchor_frame_id": "frame_003870", "shot": 6, "contexts_frame_id": ["frame_003870", "frame_000910", "frame_001520", "frame_004730", "frame_001580", "frame_002100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002236", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "13c3e046d7", "gt_frame_id": "frame_003570", "anchor_frame_id": "frame_002850", "shot": 6, "contexts_frame_id": ["frame_002850", "frame_007350", "frame_002400", "frame_007700", "frame_007620", "frame_008520"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002239", "dataset": "scannetpp", "task_type": "translation", "scene_id": "1ada7a0617", "gt_frame_id": "frame_005080", "anchor_frame_id": "frame_005490", "shot": 6, "contexts_frame_id": ["frame_005490", "frame_000990", "frame_003960", "frame_001560", "frame_000490", "frame_001460"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002242", "dataset": "scannetpp", "task_type": "translation", "scene_id": "25f3b7a318", "gt_frame_id": "frame_004100", "anchor_frame_id": "frame_002980", "shot": 6, "contexts_frame_id": ["frame_002980", "frame_005340", "frame_005400", "frame_001380", "frame_003750", "frame_002340"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.1 meters, then move backward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002245", "dataset": "scannetpp", "task_type": "composition", "scene_id": "25f3b7a318", "gt_frame_id": "frame_002470", "anchor_frame_id": "frame_001760", "shot": 6, "contexts_frame_id": ["frame_001760", "frame_001900", "frame_004210", "frame_004150", "frame_005040", "frame_004900"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002248", "dataset": "scannetpp", "task_type": "translation", "scene_id": "27dd4da69e", "gt_frame_id": "frame_002380", "anchor_frame_id": "frame_002020", "shot": 6, "contexts_frame_id": ["frame_002020", "frame_003600", "frame_004200", "frame_009170", "frame_000640", "frame_023520"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002251", "dataset": "scannetpp", "task_type": "translation", "scene_id": "27dd4da69e", "gt_frame_id": "frame_000630", "anchor_frame_id": "frame_000830", "shot": 6, "contexts_frame_id": ["frame_000830", "frame_023820", "frame_009300", "frame_005310", "frame_024600", "frame_004420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002254", "dataset": "scannetpp", "task_type": "translation", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023360", "anchor_frame_id": "frame_024000", "shot": 6, "contexts_frame_id": ["frame_024000", "frame_009230", "frame_004060", "frame_024360", "frame_024580", "frame_003500"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002257", "dataset": "scannetpp", "task_type": "composition", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005030", "anchor_frame_id": "frame_006260", "shot": 6, "contexts_frame_id": ["frame_006260", "frame_009150", "frame_023460", "frame_005290", "frame_004710", "frame_024580"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around, then move backward about 1.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002260", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "31a2c91c43", "gt_frame_id": "frame_001150", "anchor_frame_id": "frame_000040", "shot": 6, "contexts_frame_id": ["frame_000040", "frame_004890", "frame_000740", "frame_005030", "frame_005960", "frame_004630"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002263", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "31a2c91c43", "gt_frame_id": "frame_000210", "anchor_frame_id": "frame_001270", "shot": 6, "contexts_frame_id": ["frame_001270", "frame_005360", "frame_005480", "frame_005640", "frame_001020", "frame_001660"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002266", "dataset": "scannetpp", "task_type": "translation", "scene_id": "3864514494", "gt_frame_id": "frame_001030", "anchor_frame_id": "frame_002540", "shot": 6, "contexts_frame_id": ["frame_002540", "frame_002510", "frame_000020", "frame_002260", "frame_006180", "frame_000640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002269", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007010", "anchor_frame_id": "frame_007170", "shot": 6, "contexts_frame_id": ["frame_007170", "frame_006680", "frame_006740", "frame_005030", "frame_005130", "frame_004540"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002272", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007170", "anchor_frame_id": "frame_007010", "shot": 6, "contexts_frame_id": ["frame_007010", "frame_006700", "frame_014090", "frame_004970", "frame_007820", "frame_005230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002275", "dataset": "scannetpp", "task_type": "composition", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_006590", "anchor_frame_id": "frame_005300", "shot": 6, "contexts_frame_id": ["frame_005300", "frame_005130", "frame_005030", "frame_014130", "frame_004950", "frame_006330"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.7 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002278", "dataset": "scannetpp", "task_type": "composition", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_012480", "anchor_frame_id": "frame_012730", "shot": 6, "contexts_frame_id": ["frame_012730", "frame_011200", "frame_011120", "frame_012140", "frame_009730", "frame_012320"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move forward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002281", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_002320", "anchor_frame_id": "frame_002200", "shot": 6, "contexts_frame_id": ["frame_002200", "frame_002040", "frame_005750", "frame_005830", "frame_000280", "frame_000160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002284", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5748ce6f01", "gt_frame_id": "frame_002520", "anchor_frame_id": "frame_002340", "shot": 6, "contexts_frame_id": ["frame_002340", "frame_006470", "frame_000090", "frame_003110", "frame_001390", "frame_001550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002287", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5748ce6f01", "gt_frame_id": "frame_002830", "anchor_frame_id": "frame_002550", "shot": 6, "contexts_frame_id": ["frame_002550", "frame_000060", "frame_003210", "frame_004750", "frame_001570", "frame_005600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002290", "dataset": "scannetpp", "task_type": "composition", "scene_id": "578511c8a9", "gt_frame_id": "frame_001130", "anchor_frame_id": "frame_001220", "shot": 6, "contexts_frame_id": ["frame_001220", "frame_002100", "frame_000720", "frame_002120", "frame_013320", "frame_011420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move left about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002293", "dataset": "scannetpp", "task_type": "translation", "scene_id": "578511c8a9", "gt_frame_id": "frame_001090", "anchor_frame_id": "frame_000520", "shot": 6, "contexts_frame_id": ["frame_000520", "frame_002080", "frame_002000", "frame_003380", "frame_002160", "frame_011330"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002296", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5942004064", "gt_frame_id": "frame_006010", "anchor_frame_id": "frame_005590", "shot": 6, "contexts_frame_id": ["frame_005590", "frame_006200", "frame_007320", "frame_008300", "frame_008440", "frame_005160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then move forward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002299", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5942004064", "gt_frame_id": "frame_005870", "anchor_frame_id": "frame_007320", "shot": 6, "contexts_frame_id": ["frame_007320", "frame_006160", "frame_009440", "frame_007680", "frame_008260", "frame_008360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.0 meters, then move left about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002302", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5942004064", "gt_frame_id": "frame_009200", "anchor_frame_id": "frame_008800", "shot": 6, "contexts_frame_id": ["frame_008800", "frame_008440", "frame_008380", "frame_003660", "frame_008560", "frame_005560"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.6 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002305", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "5eb31827b7", "gt_frame_id": "frame_004710", "anchor_frame_id": "frame_004260", "shot": 6, "contexts_frame_id": ["frame_004260", "frame_005650", "frame_005790", "frame_004950", "frame_001580", "frame_001840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002308", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_000430", "anchor_frame_id": "frame_000500", "shot": 6, "contexts_frame_id": ["frame_000500", "frame_005620", "frame_004280", "frame_005720", "frame_006200", "frame_002140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002311", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_003480", "anchor_frame_id": "frame_003080", "shot": 6, "contexts_frame_id": ["frame_003080", "frame_002530", "frame_002110", "frame_004150", "frame_005170", "frame_006700"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 1.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002314", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_002710", "anchor_frame_id": "frame_003080", "shot": 6, "contexts_frame_id": ["frame_003080", "frame_002970", "frame_002190", "frame_002450", "frame_000600", "frame_000360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move backward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002317", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_003910", "anchor_frame_id": "frame_004950", "shot": 6, "contexts_frame_id": ["frame_004950", "frame_002110", "frame_003490", "frame_003370", "frame_005650", "frame_004510"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.4 meters, then move left about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002320", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_003810", "anchor_frame_id": "frame_003980", "shot": 6, "contexts_frame_id": ["frame_003980", "frame_002970", "frame_004350", "frame_005570", "frame_005650", "frame_005470"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.0 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002323", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_003910", "anchor_frame_id": "frame_002940", "shot": 6, "contexts_frame_id": ["frame_002940", "frame_002110", "frame_004150", "frame_005650", "frame_002270", "frame_003430"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002326", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_002740", "anchor_frame_id": "frame_003480", "shot": 6, "contexts_frame_id": ["frame_003480", "frame_004370", "frame_002950", "frame_002530", "frame_002450", "frame_000600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002329", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_003080", "anchor_frame_id": "frame_003480", "shot": 6, "contexts_frame_id": ["frame_003480", "frame_004390", "frame_002190", "frame_002450", "frame_000600", "frame_000360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.3 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002332", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_002520", "anchor_frame_id": "frame_000770", "shot": 6, "contexts_frame_id": ["frame_000770", "frame_000740", "frame_002190", "frame_002950", "frame_002110", "frame_004150"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002335", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_003070", "anchor_frame_id": "frame_002710", "shot": 6, "contexts_frame_id": ["frame_002710", "frame_002110", "frame_004150", "frame_005590", "frame_003370", "frame_002270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002338", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_001650", "anchor_frame_id": "frame_001380", "shot": 6, "contexts_frame_id": ["frame_001380", "frame_007390", "frame_001850", "frame_007230", "frame_000940", "frame_000740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002341", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_004010", "anchor_frame_id": "frame_003800", "shot": 6, "contexts_frame_id": ["frame_003800", "frame_002110", "frame_004950", "frame_003470", "frame_002490", "frame_002330"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002344", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_007970", "anchor_frame_id": "frame_007820", "shot": 6, "contexts_frame_id": ["frame_007820", "frame_001320", "frame_001660", "frame_007230", "frame_000940", "frame_000740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002347", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_002930", "anchor_frame_id": "frame_002110", "shot": 6, "contexts_frame_id": ["frame_002110", "frame_005550", "frame_003870", "frame_002190", "frame_002530", "frame_002450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.5 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002350", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_004940", "anchor_frame_id": "frame_003920", "shot": 6, "contexts_frame_id": ["frame_003920", "frame_004030", "frame_004150", "frame_005170", "frame_003530", "frame_002190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters, then move right about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002353", "dataset": "scannetpp", "task_type": "translation", "scene_id": "5f99900f09", "gt_frame_id": "frame_001660", "anchor_frame_id": "frame_001850", "shot": 6, "contexts_frame_id": ["frame_001850", "frame_007940", "frame_007390", "frame_007230", "frame_000940", "frame_000740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.0 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002356", "dataset": "scannetpp", "task_type": "composition", "scene_id": "5f99900f09", "gt_frame_id": "frame_001380", "anchor_frame_id": "frame_001650", "shot": 6, "contexts_frame_id": ["frame_001650", "frame_007230", "frame_000300", "frame_000940", "frame_000740", "frame_002570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002359", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "6115eddb86", "gt_frame_id": "frame_005550", "anchor_frame_id": "frame_005670", "shot": 6, "contexts_frame_id": ["frame_005670", "frame_003570", "frame_001790", "frame_003750", "frame_006840", "frame_007070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002362", "dataset": "scannetpp", "task_type": "translation", "scene_id": "7831862f02", "gt_frame_id": "frame_005500", "anchor_frame_id": "frame_005410", "shot": 6, "contexts_frame_id": ["frame_005410", "frame_001030", "frame_003700", "frame_008170", "frame_006580", "frame_007180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002365", "dataset": "scannetpp", "task_type": "translation", "scene_id": "7831862f02", "gt_frame_id": "frame_001600", "anchor_frame_id": "frame_001960", "shot": 6, "contexts_frame_id": ["frame_001960", "frame_005690", "frame_006400", "frame_005630", "frame_005570", "frame_005180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.5 meters, then move backward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002368", "dataset": "scannetpp", "task_type": "translation", "scene_id": "7831862f02", "gt_frame_id": "frame_001980", "anchor_frame_id": "frame_001620", "shot": 6, "contexts_frame_id": ["frame_001620", "frame_005790", "frame_001470", "frame_000950", "frame_005100", "frame_006840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.7 meters, then move forward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002371", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "825d228aec", "gt_frame_id": "frame_000610", "anchor_frame_id": "frame_000880", "shot": 6, "contexts_frame_id": ["frame_000880", "frame_000040", "frame_003050", "frame_001600", "frame_001360", "frame_001260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002374", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "825d228aec", "gt_frame_id": "frame_000870", "anchor_frame_id": "frame_000600", "shot": 6, "contexts_frame_id": ["frame_000600", "frame_002910", "frame_003410", "frame_003650", "frame_001560", "frame_001600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002377", "dataset": "scannetpp", "task_type": "translation", "scene_id": "99fa5c25e1", "gt_frame_id": "frame_007270", "anchor_frame_id": "frame_006740", "shot": 6, "contexts_frame_id": ["frame_006740", "frame_002270", "frame_000160", "frame_001170", "frame_000510", "frame_002410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002380", "dataset": "scannetpp", "task_type": "composition", "scene_id": "99fa5c25e1", "gt_frame_id": "frame_003570", "anchor_frame_id": "frame_004180", "shot": 6, "contexts_frame_id": ["frame_004180", "frame_006730", "frame_006750", "frame_002250", "frame_003270", "frame_004050"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002383", "dataset": "scannetpp", "task_type": "translation", "scene_id": "99fa5c25e1", "gt_frame_id": "frame_006770", "anchor_frame_id": "frame_007270", "shot": 6, "contexts_frame_id": ["frame_007270", "frame_000160", "frame_002250", "frame_002310", "frame_007010", "frame_000000"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002386", "dataset": "scannetpp", "task_type": "translation", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_003300", "anchor_frame_id": "frame_002220", "shot": 6, "contexts_frame_id": ["frame_002220", "frame_000340", "frame_003080", "frame_000540", "frame_002460", "frame_001780"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002389", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "acd95847c5", "gt_frame_id": "frame_007680", "anchor_frame_id": "frame_008160", "shot": 6, "contexts_frame_id": ["frame_008160", "frame_009470", "frame_007920", "frame_006640", "frame_010810", "frame_000410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002392", "dataset": "scannetpp", "task_type": "translation", "scene_id": "acd95847c5", "gt_frame_id": "frame_004900", "anchor_frame_id": "frame_004140", "shot": 6, "contexts_frame_id": ["frame_004140", "frame_004020", "frame_003880", "frame_010350", "frame_000550", "frame_006140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then move forward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002395", "dataset": "scannetpp", "task_type": "translation", "scene_id": "acd95847c5", "gt_frame_id": "frame_005370", "anchor_frame_id": "frame_004140", "shot": 6, "contexts_frame_id": ["frame_004140", "frame_004880", "frame_003860", "frame_002890", "frame_000410", "frame_006260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002398", "dataset": "scannetpp", "task_type": "translation", "scene_id": "acd95847c5", "gt_frame_id": "frame_004140", "anchor_frame_id": "frame_004900", "shot": 6, "contexts_frame_id": ["frame_004900", "frame_003980", "frame_008200", "frame_000470", "frame_000550", "frame_006140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.1 meters, then move backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002401", "dataset": "scannetpp", "task_type": "translation", "scene_id": "acd95847c5", "gt_frame_id": "frame_002850", "anchor_frame_id": "frame_001650", "shot": 6, "contexts_frame_id": ["frame_001650", "frame_003840", "frame_005320", "frame_009650", "frame_001130", "frame_000570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then move backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002404", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "b0a08200c9", "gt_frame_id": "frame_003100", "anchor_frame_id": "frame_004440", "shot": 6, "contexts_frame_id": ["frame_004440", "frame_007180", "frame_003660", "frame_002590", "frame_003580", "frame_000140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002407", "dataset": "scannetpp", "task_type": "translation", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_000830", "anchor_frame_id": "frame_000930", "shot": 6, "contexts_frame_id": ["frame_000930", "frame_004530", "frame_002210", "frame_001540", "frame_003410", "frame_003250"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002410", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002340", "anchor_frame_id": "frame_002530", "shot": 6, "contexts_frame_id": ["frame_002530", "frame_002590", "frame_004490", "frame_004010", "frame_000720", "frame_003070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002413", "dataset": "scannetpp", "task_type": "translation", "scene_id": "c4c04e6d6c", "gt_frame_id": "frame_009450", "anchor_frame_id": "frame_009230", "shot": 6, "contexts_frame_id": ["frame_009230", "frame_006400", "frame_002800", "frame_007870", "frame_007770", "frame_003850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters, then move right about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002416", "dataset": "scannetpp", "task_type": "translation", "scene_id": "c4c04e6d6c", "gt_frame_id": "frame_006430", "anchor_frame_id": "frame_006230", "shot": 6, "contexts_frame_id": ["frame_006230", "frame_002800", "frame_003340", "frame_006190", "frame_006020", "frame_004980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002419", "dataset": "scannetpp", "task_type": "composition", "scene_id": "c4c04e6d6c", "gt_frame_id": "frame_009260", "anchor_frame_id": "frame_009060", "shot": 6, "contexts_frame_id": ["frame_009060", "frame_006440", "frame_003340", "frame_002800", "frame_008490", "frame_008550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 1.0 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002422", "dataset": "scannetpp", "task_type": "translation", "scene_id": "c5439f4607", "gt_frame_id": "frame_003900", "anchor_frame_id": "frame_002480", "shot": 6, "contexts_frame_id": ["frame_002480", "frame_002000", "frame_003180", "frame_004120", "frame_004780", "frame_005280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002425", "dataset": "scannetpp", "task_type": "composition", "scene_id": "c5439f4607", "gt_frame_id": "frame_004610", "anchor_frame_id": "frame_004840", "shot": 6, "contexts_frame_id": ["frame_004840", "frame_004940", "frame_002430", "frame_002740", "frame_007220", "frame_003940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002428", "dataset": "scannetpp", "task_type": "composition", "scene_id": "cc5237fd77", "gt_frame_id": "frame_001000", "anchor_frame_id": "frame_000230", "shot": 6, "contexts_frame_id": ["frame_000230", "frame_000140", "frame_005470", "frame_002930", "frame_000840", "frame_001300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.6 meters, then turn around.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002431", "dataset": "scannetpp", "task_type": "composition", "scene_id": "cc5237fd77", "gt_frame_id": "frame_008240", "anchor_frame_id": "frame_007740", "shot": 6, "contexts_frame_id": ["frame_007740", "frame_000840", "frame_004870", "frame_000520", "frame_004990", "frame_005110"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002434", "dataset": "scannetpp", "task_type": "composition", "scene_id": "cc5237fd77", "gt_frame_id": "frame_007630", "anchor_frame_id": "frame_007740", "shot": 6, "contexts_frame_id": ["frame_007740", "frame_007340", "frame_000980", "frame_004790", "frame_008240", "frame_000420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002437", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "e7af285f7d", "gt_frame_id": "frame_006930", "anchor_frame_id": "frame_008500", "shot": 6, "contexts_frame_id": ["frame_008500", "frame_007980", "frame_000940", "frame_007460", "frame_000480", "frame_007360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002440", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "e7af285f7d", "gt_frame_id": "frame_008500", "anchor_frame_id": "frame_006930", "shot": 6, "contexts_frame_id": ["frame_006930", "frame_000500", "frame_007640", "frame_000940", "frame_008960", "frame_009180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002443", "dataset": "scannetpp", "task_type": "translation", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_000780", "anchor_frame_id": "frame_001360", "shot": 6, "contexts_frame_id": ["frame_001360", "frame_004980", "frame_001080", "frame_000300", "frame_008370", "frame_002780"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.7 meters, then move right about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002446", "dataset": "scannetpp", "task_type": "translation", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_001080", "anchor_frame_id": "frame_001360", "shot": 6, "contexts_frame_id": ["frame_001360", "frame_000580", "frame_000540", "frame_008390", "frame_002740", "frame_008150"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002449", "dataset": "scannetpp", "task_type": "translation", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_002780", "anchor_frame_id": "frame_003180", "shot": 6, "contexts_frame_id": ["frame_003180", "frame_003120", "frame_002520", "frame_000280", "frame_000780", "frame_006580"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002452", "dataset": "scannetpp", "task_type": "composition", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_002520", "anchor_frame_id": "frame_001670", "shot": 6, "contexts_frame_id": ["frame_001670", "frame_000280", "frame_008250", "frame_008310", "frame_002780", "frame_007200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move left about 1.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002455", "dataset": "scannetpp", "task_type": "translation", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_000780", "anchor_frame_id": "frame_001070", "shot": 6, "contexts_frame_id": ["frame_001070", "frame_001140", "frame_004980", "frame_000280", "frame_008430", "frame_003060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.9 meters, then move right about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002458", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_000560", "anchor_frame_id": "frame_000280", "shot": 6, "contexts_frame_id": ["frame_000280", "frame_001080", "frame_001140", "frame_000860", "frame_008430", "frame_003060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002461", "dataset": "scannetpp", "task_type": "rotation", "scene_id": "f9f95681fd", "gt_frame_id": "frame_007270", "anchor_frame_id": "frame_006450", "shot": 6, "contexts_frame_id": ["frame_006450", "frame_004300", "frame_000060", "frame_000200", "frame_001160", "frame_002870"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002464", "dataset": "scannetpp", "task_type": "composition", "scene_id": "fb5a96b1a2", "gt_frame_id": "frame_004420", "anchor_frame_id": "frame_003280", "shot": 6, "contexts_frame_id": ["frame_003280", "frame_008250", "frame_000940", "frame_005190", "frame_005610", "frame_007510"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 1.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002467", "dataset": "scannetpp", "task_type": "translation", "scene_id": "fb5a96b1a2", "gt_frame_id": "frame_004170", "anchor_frame_id": "frame_004430", "shot": 6, "contexts_frame_id": ["frame_004430", "frame_000880", "frame_000400", "frame_003180", "frame_005370", "frame_005290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002470", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "09c1414f1b", "gt_frame_id": "frame_003510", "anchor_frame_id": "frame_003020", "shot": 6, "contexts_frame_id": ["frame_003020", "frame_002380", "frame_003980", "frame_004630", "frame_004740", "frame_004880"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the coffee table, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the coffee table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002473", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "09c1414f1b", "gt_frame_id": "frame_000640", "anchor_frame_id": "frame_000280", "shot": 6, "contexts_frame_id": ["frame_000280", "frame_001750", "frame_000100", "frame_002160", "frame_004740", "frame_004880"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002476", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "09c1414f1b", "gt_frame_id": "frame_001520", "anchor_frame_id": "frame_000280", "shot": 6, "contexts_frame_id": ["frame_000280", "frame_006100", "frame_000760", "frame_001830", "frame_003500", "frame_003940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002479", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "09c1414f1b", "gt_frame_id": "frame_000490", "anchor_frame_id": "frame_001330", "shot": 6, "contexts_frame_id": ["frame_001330", "frame_000300", "frame_006100", "frame_002160", "frame_004740", "frame_004880"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the plant, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the plant, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002482", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "0d2ee665be", "gt_frame_id": "frame_001180", "anchor_frame_id": "frame_001540", "shot": 6, "contexts_frame_id": ["frame_001540", "frame_003640", "frame_001640", "frame_001820", "frame_000910", "frame_000690"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002485", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "1ada7a0617", "gt_frame_id": "frame_002120", "anchor_frame_id": "frame_001580", "shot": 6, "contexts_frame_id": ["frame_001580", "frame_004260", "frame_002720", "frame_002760", "frame_005140", "frame_005500"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the blinds, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the blinds, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002488", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "21d970d8de", "gt_frame_id": "frame_005700", "anchor_frame_id": "frame_001210", "shot": 6, "contexts_frame_id": ["frame_001210", "frame_003720", "frame_006010", "frame_004040", "frame_002780", "frame_007670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the door, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the door, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002491", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023580", "anchor_frame_id": "frame_005270", "shot": 6, "contexts_frame_id": ["frame_005270", "frame_024620", "frame_002620", "frame_024280", "frame_009300", "frame_009670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the kitchen cabinet, which is near the center of the first image, about 90 degrees to the left around the object. Keep a similar distance from the kitchen cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002494", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005210", "anchor_frame_id": "frame_000790", "shot": 6, "contexts_frame_id": ["frame_000790", "frame_004930", "frame_023840", "frame_023280", "frame_024240", "frame_002620"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the range hood, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the range hood, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002497", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "27dd4da69e", "gt_frame_id": "frame_000790", "anchor_frame_id": "frame_005210", "shot": 6, "contexts_frame_id": ["frame_005210", "frame_005270", "frame_005050", "frame_002180", "frame_006530", "frame_023900"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the range hood, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the range hood, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002500", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_004260", "anchor_frame_id": "frame_006010", "shot": 6, "contexts_frame_id": ["frame_006010", "frame_003080", "frame_013890", "frame_001990", "frame_002370", "frame_002410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the drying rack, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the drying rack, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002503", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5748ce6f01", "gt_frame_id": "frame_003860", "anchor_frame_id": "frame_005920", "shot": 6, "contexts_frame_id": ["frame_005920", "frame_002010", "frame_000770", "frame_000090", "frame_001710", "frame_001550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002506", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5748ce6f01", "gt_frame_id": "frame_003930", "anchor_frame_id": "frame_000370", "shot": 6, "contexts_frame_id": ["frame_000370", "frame_000090", "frame_001710", "frame_001550", "frame_002070", "frame_002230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 90 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002509", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5748ce6f01", "gt_frame_id": "frame_002180", "anchor_frame_id": "frame_000370", "shot": 6, "contexts_frame_id": ["frame_000370", "frame_003390", "frame_001970", "frame_000090", "frame_001710", "frame_001550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002512", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5942004064", "gt_frame_id": "frame_009320", "anchor_frame_id": "frame_009390", "shot": 6, "contexts_frame_id": ["frame_009390", "frame_005940", "frame_007340", "frame_008540", "frame_008740", "frame_008800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002515", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5942004064", "gt_frame_id": "frame_009730", "anchor_frame_id": "frame_009300", "shot": 6, "contexts_frame_id": ["frame_009300", "frame_005840", "frame_003740", "frame_003830", "frame_005740", "frame_008180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002518", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5942004064", "gt_frame_id": "frame_005960", "anchor_frame_id": "frame_006090", "shot": 6, "contexts_frame_id": ["frame_006090", "frame_007360", "frame_005680", "frame_008320", "frame_008380", "frame_005060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002521", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_003800", "anchor_frame_id": "frame_000490", "shot": 6, "contexts_frame_id": ["frame_000490", "frame_005940", "frame_004440", "frame_003450", "frame_001460", "frame_000280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the copy machine, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the copy machine, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002524", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_006100", "anchor_frame_id": "frame_006220", "shot": 6, "contexts_frame_id": ["frame_006220", "frame_007110", "frame_001850", "frame_006360", "frame_006700", "frame_005170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002527", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_007370", "anchor_frame_id": "frame_007560", "shot": 6, "contexts_frame_id": ["frame_007560", "frame_001440", "frame_005900", "frame_008120", "frame_006460", "frame_005270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002530", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_007570", "anchor_frame_id": "frame_007370", "shot": 6, "contexts_frame_id": ["frame_007370", "frame_001660", "frame_001800", "frame_008120", "frame_005250", "frame_005170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002533", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_007580", "anchor_frame_id": "frame_007460", "shot": 6, "contexts_frame_id": ["frame_007460", "frame_001660", "frame_001740", "frame_008120", "frame_005250", "frame_005170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002536", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_000540", "anchor_frame_id": "frame_000870", "shot": 6, "contexts_frame_id": ["frame_000870", "frame_000160", "frame_001100", "frame_002450", "frame_002190", "frame_002950"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002539", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_000870", "anchor_frame_id": "frame_000560", "shot": 6, "contexts_frame_id": ["frame_000560", "frame_002450", "frame_000120", "frame_001540", "frame_001440", "frame_001660"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002542", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_000340", "anchor_frame_id": "frame_000560", "shot": 6, "contexts_frame_id": ["frame_000560", "frame_001220", "frame_000060", "frame_002330", "frame_002470", "frame_003130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002545", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "5f99900f09", "gt_frame_id": "frame_004740", "anchor_frame_id": "frame_004510", "shot": 6, "contexts_frame_id": ["frame_004510", "frame_005650", "frame_004150", "frame_005170", "frame_005270", "frame_006400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002548", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "7831862f02", "gt_frame_id": "frame_002070", "anchor_frame_id": "frame_007900", "shot": 6, "contexts_frame_id": ["frame_007900", "frame_000950", "frame_005960", "frame_007100", "frame_004410", "frame_000120"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002551", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "7831862f02", "gt_frame_id": "frame_005530", "anchor_frame_id": "frame_005990", "shot": 6, "contexts_frame_id": ["frame_005990", "frame_005160", "frame_001250", "frame_008070", "frame_003080", "frame_007040"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bench, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the bench, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002554", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "7831862f02", "gt_frame_id": "frame_001080", "anchor_frame_id": "frame_005520", "shot": 6, "contexts_frame_id": ["frame_005520", "frame_005100", "frame_008050", "frame_006200", "frame_007100", "frame_002540"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002557", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "7831862f02", "gt_frame_id": "frame_007150", "anchor_frame_id": "frame_002370", "shot": 6, "contexts_frame_id": ["frame_002370", "frame_003550", "frame_006460", "frame_008110", "frame_002720", "frame_005060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the box, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the box, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002560", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "7b6477cb95", "gt_frame_id": "frame_003580", "anchor_frame_id": "frame_005910", "shot": 6, "contexts_frame_id": ["frame_005910", "frame_003150", "frame_003340", "frame_005850", "frame_005700", "frame_006150"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the office chair, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the office chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002563", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "9071e139d9", "gt_frame_id": "frame_004240", "anchor_frame_id": "frame_003840", "shot": 6, "contexts_frame_id": ["frame_003840", "frame_000210", "frame_000310", "frame_003190", "frame_000660", "frame_000820"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002566", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "9071e139d9", "gt_frame_id": "frame_003840", "anchor_frame_id": "frame_004270", "shot": 6, "contexts_frame_id": ["frame_004270", "frame_004360", "frame_000150", "frame_001420", "frame_000960", "frame_000840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002569", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_002220", "anchor_frame_id": "frame_000190", "shot": 6, "contexts_frame_id": ["frame_000190", "frame_001780", "frame_001680", "frame_001940", "frame_000400", "frame_000540"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the monitor, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the monitor, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002572", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_000190", "anchor_frame_id": "frame_002220", "shot": 6, "contexts_frame_id": ["frame_002220", "frame_001760", "frame_001900", "frame_002280", "frame_003160", "frame_003620"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the monitor, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the monitor, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002575", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "a8bf42d646", "gt_frame_id": "frame_004720", "anchor_frame_id": "frame_005100", "shot": 6, "contexts_frame_id": ["frame_005100", "frame_005260", "frame_005820", "frame_006920", "frame_004480", "frame_004460"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002578", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "a8bf42d646", "gt_frame_id": "frame_004960", "anchor_frame_id": "frame_001020", "shot": 6, "contexts_frame_id": ["frame_001020", "frame_000820", "frame_004780", "frame_004740", "frame_004600", "frame_007420"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the door, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the door, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002581", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "a8bf42d646", "gt_frame_id": "frame_001020", "anchor_frame_id": "frame_004960", "shot": 6, "contexts_frame_id": ["frame_004960", "frame_004460", "frame_000740", "frame_002140", "frame_002200", "frame_003780"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the door, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the door, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002584", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "bcd2436daf", "gt_frame_id": "frame_004410", "anchor_frame_id": "frame_004910", "shot": 6, "contexts_frame_id": ["frame_004910", "frame_002070", "frame_001630", "frame_002320", "frame_001370", "frame_000080"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 60 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002587", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "bcd2436daf", "gt_frame_id": "frame_004380", "anchor_frame_id": "frame_004910", "shot": 6, "contexts_frame_id": ["frame_004910", "frame_000140", "frame_000300", "frame_002090", "frame_002220", "frame_002280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the pillow, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the pillow, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002590", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002620", "anchor_frame_id": "frame_004060", "shot": 6, "contexts_frame_id": ["frame_004060", "frame_002290", "frame_000740", "frame_003390", "frame_003610", "frame_000240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the heater, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the heater, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002593", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_004810", "anchor_frame_id": "frame_008360", "shot": 6, "contexts_frame_id": ["frame_008360", "frame_007980", "frame_004470", "frame_007560", "frame_005890", "frame_001580"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002596", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_005250", "anchor_frame_id": "frame_005040", "shot": 6, "contexts_frame_id": ["frame_005040", "frame_000360", "frame_000200", "frame_005470", "frame_004470", "frame_007840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the desk, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the desk, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002599", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_005620", "anchor_frame_id": "frame_000160", "shot": 6, "contexts_frame_id": ["frame_000160", "frame_001540", "frame_001640", "frame_000260", "frame_008100", "frame_006570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002602", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_000160", "anchor_frame_id": "frame_005620", "shot": 6, "contexts_frame_id": ["frame_005620", "frame_005910", "frame_006150", "frame_002970", "frame_004490", "frame_006810"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the bookshelf, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the bookshelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002605", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_005360", "anchor_frame_id": "frame_005000", "shot": 6, "contexts_frame_id": ["frame_005000", "frame_000280", "frame_008120", "frame_004510", "frame_006810", "frame_000960"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the monitor, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the monitor, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002608", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_005000", "anchor_frame_id": "frame_005370", "shot": 6, "contexts_frame_id": ["frame_005370", "frame_000280", "frame_001620", "frame_006330", "frame_007980", "frame_003030"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the monitor, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the monitor, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002611", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "cc5237fd77", "gt_frame_id": "frame_005350", "anchor_frame_id": "frame_000390", "shot": 6, "contexts_frame_id": ["frame_000390", "frame_000300", "frame_008120", "frame_004530", "frame_004470", "frame_007840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the monitor, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the monitor, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002614", "dataset": "scannetpp", "task_type": "anchor_guided_relative_repositioning", "scene_id": "fb5a96b1a2", "gt_frame_id": "frame_002440", "anchor_frame_id": "frame_002870", "shot": 6, "contexts_frame_id": ["frame_002870", "frame_006840", "frame_000780", "frame_003900", "frame_003860", "frame_005790"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002617", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "09c1414f1b", "gt_frame_id": "frame_002560", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001830", "frame_003000", "frame_003880", "frame_003980", "frame_004610", "frame_004740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe coffee table behind the footstool.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002620", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "09c1414f1b", "gt_frame_id": "frame_007480", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007890", "frame_008190", "frame_007750", "frame_008450", "frame_008550", "frame_005800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe kitchen cabinet in front of the window.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 3.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002623", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "0d2ee665be", "gt_frame_id": "frame_001540", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002120", "frame_003680", "frame_001190", "frame_003840", "frame_000930", "frame_004020"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table under the laptop, under the monitor, and to the right of the office chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002626", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "0d2ee665be", "gt_frame_id": "frame_001920", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000910", "frame_000970", "frame_003640", "frame_003680", "frame_004770", "frame_004750"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe laptop on the table, and to the left of the book shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002629", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "25f3b7a318", "gt_frame_id": "frame_000310", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002500", "frame_002760", "frame_004070", "frame_003010", "frame_005360", "frame_000860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe pedestal fan to the left of the backpack, and to the left of the laundry basket.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002632", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "25f3b7a318", "gt_frame_id": "frame_000160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006240", "frame_001780", "frame_001720", "frame_001640", "frame_000800", "frame_005400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe blanket to the right of the laundry basket, and to the right of the backpack.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002635", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005110", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000060", "frame_009150", "frame_001710", "frame_003920", "frame_002860", "frame_024080"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe clothes dryer to the left of the wok pan, and to the left of the microwave.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002638", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "286b55a2bf", "gt_frame_id": "frame_004610", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002820", "frame_004080", "frame_002980", "frame_001060", "frame_001000", "frame_000880"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe sink under the tap.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002641", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "31a2c91c43", "gt_frame_id": "frame_004300", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005360", "frame_005480", "frame_005800", "frame_003240", "frame_003180", "frame_002040"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe towel warmer to the right of the jeans.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002644", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "31a2c91c43", "gt_frame_id": "frame_004310", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005480", "frame_005800", "frame_003240", "frame_003180", "frame_002240", "frame_002060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe towel to the right of two jeanses.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002647", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "38d58a7a31", "gt_frame_id": "frame_009870", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001360", "frame_012170", "frame_006010", "frame_005170", "frame_002200", "frame_012120"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe blinds on the window sill.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002650", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "38d58a7a31", "gt_frame_id": "frame_008640", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006490", "frame_011520", "frame_002510", "frame_009490", "frame_008310", "frame_005990"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe coat hanger to the right of three chairs.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002653", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_009770", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_013370", "frame_012040", "frame_010590", "frame_012120", "frame_012240", "frame_012360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the left of the air conditioner.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002656", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_003240", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005950", "frame_008000", "frame_004350", "frame_004580", "frame_004090", "frame_007400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table to the left of the chair, and behind the sofa.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002659", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_011220", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008340", "frame_011620", "frame_012880", "frame_012730", "frame_012800", "frame_010850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe piano under the piano book.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002662", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007830", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_013430", "frame_006720", "frame_007140", "frame_004950", "frame_007560", "frame_006500"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bag under the bottles.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002665", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_006950", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004110", "frame_005230", "frame_005050", "frame_007200", "frame_006700", "frame_006740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe stove on the oven, and to the right of the cutting board.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002668", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "3f15a9266d", "gt_frame_id": "frame_001110", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000900", "frame_002760", "frame_001320", "frame_001500", "frame_001620", "frame_001840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe door to the left of the bookshelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002671", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "578511c8a9", "gt_frame_id": "frame_000260", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001380", "frame_000540", "frame_001280", "frame_011330", "frame_011370", "frame_013100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table lamp on the table, and to the left of two whiteboards.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002674", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5942004064", "gt_frame_id": "frame_009770", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008440", "frame_008480", "frame_009060", "frame_004580", "frame_003830", "frame_006520"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe armchair to the left of the plant, and in front of the pegboard.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002677", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5942004064", "gt_frame_id": "frame_006510", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003810", "frame_004520", "frame_004460", "frame_009770", "frame_009060", "frame_008480"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table mat on the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002680", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5942004064", "gt_frame_id": "frame_006650", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005380", "frame_008440", "frame_008480", "frame_009060", "frame_004560", "frame_004120"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe pegboard under the picture.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002683", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5942004064", "gt_frame_id": "frame_006590", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005440", "frame_009160", "frame_008780", "frame_004140", "frame_004500", "frame_003810"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe lamp to the right of the photo frame.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002686", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5942004064", "gt_frame_id": "frame_004410", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008440", "frame_008480", "frame_009060", "frame_009790", "frame_004120", "frame_003830"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe plant to the right of the armchair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002689", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5eb31827b7", "gt_frame_id": "frame_007160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006070", "frame_005570", "frame_005530", "frame_005150", "frame_001860", "frame_007740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe bookshelf to the left of the trash can.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002692", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_005820", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004440", "frame_003790", "frame_001360", "frame_006240", "frame_007050", "frame_005260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table under the microwave, and to the right of the trash bin.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002695", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "5f99900f09", "gt_frame_id": "frame_006220", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007700", "frame_005900", "frame_007070", "frame_007230", "frame_000940", "frame_000740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table between two chairs, and to the right of the arm chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002698", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "7831862f02", "gt_frame_id": "frame_008140", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008380", "frame_002640", "frame_004250", "frame_000020", "frame_003500", "frame_002440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe box to the left of the stool, and in front of the beanbag.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002701", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_002380", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001780", "frame_000160", "frame_000300", "frame_000400", "frame_000540", "frame_000600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table below the monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002704", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "a8bf42d646", "gt_frame_id": "frame_001770", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000740", "frame_001240", "frame_004960", "frame_000200", "frame_004480", "frame_007020"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe machine to the right of the trash can.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002707", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "ac48a9b736", "gt_frame_id": "frame_008990", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_010400", "frame_010340", "frame_008600", "frame_007470", "frame_007270", "frame_007670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe shelf under the box, to the right of the table, and to the right of the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 3.2 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002710", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "acd95847c5", "gt_frame_id": "frame_002730", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000570", "frame_000430", "frame_001710", "frame_001310", "frame_004820", "frame_003980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the left of the table, and to the left of the monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002713", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "acd95847c5", "gt_frame_id": "frame_000260", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008640", "frame_008600", "frame_007380", "frame_002870", "frame_000080", "frame_007640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe monitor on the table, and to the right of the telephone.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.3 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002716", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "b0a08200c9", "gt_frame_id": "frame_002740", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006890", "frame_008560", "frame_000660", "frame_002160", "frame_003620", "frame_003160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe box to the left of the computer tower.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002719", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "c5439f4607", "gt_frame_id": "frame_001010", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008040", "frame_008100", "frame_008200", "frame_008360", "frame_009660", "frame_000860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe wooden stick to the right of the tool organizer.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002722", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "cc5237fd77", "gt_frame_id": "frame_000470", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006810", "frame_008240", "frame_004570", "frame_005210", "frame_005430", "frame_000260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe blinds under the window.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002725", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "e7af285f7d", "gt_frame_id": "frame_000120", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005040", "frame_004440", "frame_003400", "frame_000360", "frame_002060", "frame_001840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe mirror behind the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002728", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_001990", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007200", "frame_002520", "frame_002980", "frame_000540", "frame_001660", "frame_007350"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe storage cabinet to the left of the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002731", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_008810", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006580", "frame_000780", "frame_000300", "frame_000420", "frame_008330", "frame_002780"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair behind the box.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002734", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_002680", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000280", "frame_002500", "frame_003040", "frame_003080", "frame_003140", "frame_010430"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the microwave, and to the left of the coffee machine.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002737", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f9f95681fd", "gt_frame_id": "frame_002470", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004300", "frame_006460", "frame_000060", "frame_006200", "frame_000260", "frame_007180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe plate to the left of two knife sets.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002740", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f9f95681fd", "gt_frame_id": "frame_002930", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004830", "frame_006360", "frame_000120", "frame_000240", "frame_000320", "frame_007300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe dishwasher under the stove.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002743", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f9f95681fd", "gt_frame_id": "frame_002730", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000800", "frame_000200", "frame_005320", "frame_005120", "frame_006420", "frame_004910"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe coffee machine to the left of the stove.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.6 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002746", "dataset": "scannetpp", "task_type": "obj_loc", "scene_id": "f9f95681fd", "gt_frame_id": "frame_006630", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000740", "frame_003940", "frame_007400", "frame_004440", "frame_004870", "frame_004580"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe dishwasher to the left of the kitchen towel.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002749", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "0d2ee665be", "gt_frame_id": "frame_004740", "anchor_frame_id": "frame_004560", "shot": 6, "contexts_frame_id": ["frame_004560", "frame_002120", "frame_003680", "frame_003620", "frame_001190", "frame_000950"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the mousepad.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002752", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "13c3e046d7", "gt_frame_id": "frame_002960", "anchor_frame_id": "frame_003580", "shot": 6, "contexts_frame_id": ["frame_003580", "frame_002340", "frame_007350", "frame_003110", "frame_006610", "frame_008730"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the billboard.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002755", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "25f3b7a318", "gt_frame_id": "frame_001560", "anchor_frame_id": "frame_001650", "shot": 6, "contexts_frame_id": ["frame_001650", "frame_005020", "frame_000940", "frame_005360", "frame_003010", "frame_004070"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the laundry basket.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002758", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005090", "anchor_frame_id": "frame_002690", "shot": 6, "contexts_frame_id": ["frame_002690", "frame_002420", "frame_002640", "frame_004670", "frame_000060", "frame_009170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the washing machine.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002761", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005150", "anchor_frame_id": "frame_002100", "shot": 6, "contexts_frame_id": ["frame_002100", "frame_006310", "frame_023260", "frame_005910", "frame_025570", "frame_000220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the electric stove.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002764", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "27dd4da69e", "gt_frame_id": "frame_006260", "anchor_frame_id": "frame_005610", "shot": 6, "contexts_frame_id": ["frame_005610", "frame_004710", "frame_009650", "frame_005930", "frame_004220", "frame_003920"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the window.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002767", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023940", "anchor_frame_id": "frame_023670", "shot": 6, "contexts_frame_id": ["frame_023670", "frame_004850", "frame_004990", "frame_023380", "frame_023280", "frame_003960"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the window.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002770", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "286b55a2bf", "gt_frame_id": "frame_004490", "anchor_frame_id": "frame_000910", "shot": 6, "contexts_frame_id": ["frame_000910", "frame_000880", "frame_000840", "frame_003960", "frame_001000", "frame_001480"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the hairdryer.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002773", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "31a2c91c43", "gt_frame_id": "frame_004450", "anchor_frame_id": "frame_006390", "shot": 6, "contexts_frame_id": ["frame_006390", "frame_000200", "frame_004690", "frame_001960", "frame_003100", "frame_007590"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trousers.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002776", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "3864514494", "gt_frame_id": "frame_005830", "anchor_frame_id": "frame_002990", "shot": 6, "contexts_frame_id": ["frame_002990", "frame_005430", "frame_000020", "frame_003250", "frame_000100", "frame_003630"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the espresso machine.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002779", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007140", "anchor_frame_id": "frame_007000", "shot": 6, "contexts_frame_id": ["frame_007000", "frame_012040", "frame_006700", "frame_004970", "frame_007820", "frame_005230"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trash bin near the kettle.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002782", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007560", "anchor_frame_id": "frame_006950", "shot": 6, "contexts_frame_id": ["frame_006950", "frame_006740", "frame_004950", "frame_005030", "frame_014190", "frame_005130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the printer.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002785", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_007110", "anchor_frame_id": "frame_007190", "shot": 6, "contexts_frame_id": ["frame_007190", "frame_014110", "frame_006660", "frame_007560", "frame_004950", "frame_006740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bowl near the kettle.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002788", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_006950", "anchor_frame_id": "frame_007560", "shot": 6, "contexts_frame_id": ["frame_007560", "frame_005230", "frame_004110", "frame_004970", "frame_006700", "frame_006740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the stove.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002791", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_001170", "anchor_frame_id": "frame_001770", "shot": 6, "contexts_frame_id": ["frame_001770", "frame_005170", "frame_005110", "frame_000920", "frame_006390", "frame_006010"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trash bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002794", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "5942004064", "gt_frame_id": "frame_008340", "anchor_frame_id": "frame_007870", "shot": 6, "contexts_frame_id": ["frame_007870", "frame_005700", "frame_009400", "frame_003740", "frame_008720", "frame_009180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the pillow near the pot.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002797", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_001580", "anchor_frame_id": "frame_003790", "shot": 6, "contexts_frame_id": ["frame_003790", "frame_003830", "frame_001200", "frame_000340", "frame_007050", "frame_002950"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the microwave.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002800", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "6115eddb86", "gt_frame_id": "frame_000050", "anchor_frame_id": "frame_000750", "shot": 6, "contexts_frame_id": ["frame_000750", "frame_002530", "frame_008640", "frame_008620", "frame_008520", "frame_001610"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the standing lamp.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002803", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "6115eddb86", "gt_frame_id": "frame_007780", "anchor_frame_id": "frame_008060", "shot": 6, "contexts_frame_id": ["frame_008060", "frame_008660", "frame_008150", "frame_002050", "frame_001910", "frame_001870"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the cabinet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002806", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "7bc286c1b6", "gt_frame_id": "frame_004200", "anchor_frame_id": "frame_001300", "shot": 6, "contexts_frame_id": ["frame_001300", "frame_002780", "frame_003360", "frame_000250", "frame_000430", "frame_000370"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the surface cleaning liquid.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002809", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_003630", "anchor_frame_id": "frame_002040", "shot": 6, "contexts_frame_id": ["frame_002040", "frame_001760", "frame_002220", "frame_003100", "frame_003440", "frame_000600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the folding sofa.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002812", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "acd95847c5", "gt_frame_id": "frame_001350", "anchor_frame_id": "frame_000960", "shot": 6, "contexts_frame_id": ["frame_000960", "frame_000550", "frame_005000", "frame_007540", "frame_007660", "frame_002730"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the laptop.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002815", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_000970", "anchor_frame_id": "frame_004160", "shot": 6, "contexts_frame_id": ["frame_004160", "frame_002410", "frame_002290", "frame_002890", "frame_000660", "frame_003390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the monitor near the trash can.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002818", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_003410", "anchor_frame_id": "frame_001990", "shot": 6, "contexts_frame_id": ["frame_001990", "frame_001400", "frame_003650", "frame_003870", "frame_000840", "frame_002630"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the bedside cabinet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002821", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "c50d2d1d42", "gt_frame_id": "frame_003610", "anchor_frame_id": "frame_009320", "shot": 6, "contexts_frame_id": ["frame_009320", "frame_001460", "frame_007580", "frame_007280", "frame_008390", "frame_006320"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the carboard box.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002824", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "f2dc06b1d2", "gt_frame_id": "frame_001230", "anchor_frame_id": "frame_001470", "shot": 6, "contexts_frame_id": ["frame_001470", "frame_002790", "frame_003770", "frame_001670", "frame_000330", "frame_000390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the trash bin.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002827", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_000570", "anchor_frame_id": "frame_000240", "shot": 6, "contexts_frame_id": ["frame_000240", "frame_000320", "frame_001700", "frame_001080", "frame_000860", "frame_003360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the blinds.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002830", "dataset": "scannetpp", "task_type": "loc-re-facing", "scene_id": "f9f95681fd", "gt_frame_id": "frame_005790", "anchor_frame_id": "frame_002860", "shot": 6, "contexts_frame_id": ["frame_002860", "frame_003340", "frame_006920", "frame_003800", "frame_007180", "frame_000200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the kettle.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002833", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "09c1414f1b", "gt_frame_id": "frame_001910", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002100", "frame_002380", "frame_002740", "frame_001710", "frame_001730", "frame_000380"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the plant, facing the footstool.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002836", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "0d2ee665be", "gt_frame_id": "frame_000930", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000750", "frame_000710", "frame_000690", "frame_003840", "frame_001290", "frame_001110"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFacing the nightstand, I have the cabinet behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002839", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "0d2ee665be", "gt_frame_id": "frame_000960", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001310", "frame_001350", "frame_000690", "frame_000710", "frame_003680", "frame_003640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am at a position where the cabinet is behind me and the office chair is in front of me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002842", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "0d2ee665be", "gt_frame_id": "frame_003660", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002260", "frame_001980", "frame_001580", "frame_001390", "frame_000950", "frame_001110"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the trash can, and the shelf is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002845", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "0d2ee665be", "gt_frame_id": "frame_002950", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003260", "frame_002340", "frame_002680", "frame_002760", "frame_002620", "frame_003490"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the kitchen cabinet to my left, I am facing the salad spinner.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002848", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "0d2ee665be", "gt_frame_id": "frame_004930", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000610", "frame_000140", "frame_000550", "frame_000000", "frame_002100", "frame_002260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the cabinet, facing the bag, and I can see the laundry hamper on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002851", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "1ada7a0617", "gt_frame_id": "frame_002390", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002700", "frame_003490", "frame_002110", "frame_002630", "frame_004540", "frame_004800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the office chair, I face the telephone, and I can see the monitor on the left side of the view, and the mouse near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002854", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "25f3b7a318", "gt_frame_id": "frame_002690", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004700", "frame_006480", "frame_001920", "frame_002380", "frame_003950", "frame_003770"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the carpet to my left, I am facing the headphone.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002857", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "25f3b7a318", "gt_frame_id": "frame_002750", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007380", "frame_001880", "frame_002500", "frame_004210", "frame_003670", "frame_003970"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the speaker, with the carpet to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002860", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "25f3b7a318", "gt_frame_id": "frame_003300", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002160", "frame_003670", "frame_004090", "frame_005340", "frame_005380", "frame_005440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the carpet to my left, I face the shelf.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002863", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "25f3b7a318", "gt_frame_id": "frame_004570", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001960", "frame_001940", "frame_001900", "frame_006080", "frame_005540", "frame_006940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nThe carpet is behind me as I face the shoe rack.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002866", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "25f3b7a318", "gt_frame_id": "frame_004180", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001940", "frame_002600", "frame_002380", "frame_002280", "frame_002760", "frame_003390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the headphone, and the carpet remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002869", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_000280", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_009150", "frame_009170", "frame_023320", "frame_023440", "frame_004910", "frame_009670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the laundry basket, facing the tray, and I can see the range hood on the right side of the view, and the kitchen brush near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002872", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_000990", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002420", "frame_002600", "frame_005070", "frame_000480", "frame_023900", "frame_005190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI stand with the electric pot behind me and face the wok pan.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002875", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005760", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002360", "frame_024500", "frame_005050", "frame_009320", "frame_000320", "frame_023900"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the dish drying rack to my left, I am facing the mug.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002878", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_002670", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_024500", "frame_002320", "frame_004000", "frame_024080", "frame_002260", "frame_003360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the bar stool, facing the paper towel, and I can see the bag on the left side of the view, and the paper towel package on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002881", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005320", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006970", "frame_005030", "frame_009300", "frame_023800", "frame_003120", "frame_006270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the flour bag, the bar stool is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002884", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_006260", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003900", "frame_023240", "frame_009270", "frame_005090", "frame_023800", "frame_025570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the washing machine, I am facing the window.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002887", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_025630", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004870", "frame_005350", "frame_005190", "frame_005910", "frame_023280", "frame_024340"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the clothes dryer, I face the mug, and I can see the tray on the left side of the view, and the wok pan on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002890", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005080", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_024500", "frame_004710", "frame_000060", "frame_009150", "frame_024060", "frame_002980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the washing machine, and the shelf is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002893", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023810", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_024360", "frame_023300", "frame_023320", "frame_009150", "frame_023440", "frame_005190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the laundry basket, facing the wok pan.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002896", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_002920", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_023740", "frame_009150", "frame_023300", "frame_003900", "frame_002100", "frame_002280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the electric pot to my right, I am facing the paper towel.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002899", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_004880", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_024560", "frame_024240", "frame_023280", "frame_009150", "frame_000100", "frame_006970"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the dish drying rack, with the bar stool to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002902", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023940", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003920", "frame_023300", "frame_009150", "frame_023380", "frame_023280", "frame_003600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the laundry basket, I face the window, and I can see the fridge on the left side of the view, and the shelf near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002905", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_000320", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005070", "frame_009150", "frame_023320", "frame_023280", "frame_023460", "frame_005190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the laundry basket, I am facing the oven.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002908", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_002600", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002100", "frame_003960", "frame_023240", "frame_003520", "frame_004670", "frame_005030"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the book, facing the paper towel, with the trash can to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002911", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023700", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_009670", "frame_000100", "frame_006970", "frame_005010", "frame_004670", "frame_003460"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nMy view is toward the pan, and the laundry basket is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002914", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005800", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002360", "frame_024500", "frame_005050", "frame_000020", "frame_000260", "frame_009210"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the dish drying rack, facing the flour bag, with the pan to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002917", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_024200", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005030", "frame_002620", "frame_024460", "frame_004280", "frame_006270", "frame_002100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the bag, with the clothes dryer to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002920", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_005110", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_024540", "frame_024260", "frame_023340", "frame_006330", "frame_009150", "frame_000000"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the bar stool, I face the clothes dryer, and I can see the electric stove on the right side of the view, and the sink near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002923", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_024370", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000460", "frame_023280", "frame_023240", "frame_006310", "frame_003760", "frame_002600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the bar stool, I face the jar, and I can see the paper towel package on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002926", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_023840", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_009150", "frame_023300", "frame_023280", "frame_023380", "frame_023460", "frame_000240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the range hood, the laundry basket is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002929", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_025430", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005190", "frame_000300", "frame_009170", "frame_005070", "frame_002600", "frame_002480"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the electric stove, I am facing the soap dispenser, and I can see the box on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002932", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_006300", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005030", "frame_000400", "frame_023320", "frame_024020", "frame_003000", "frame_024080"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the washing machine, I am facing the trash can, and I can see the paper towel on the right side of the view, and the window near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002935", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "27dd4da69e", "gt_frame_id": "frame_010670", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_021880", "frame_008710", "frame_008400", "frame_009870", "frame_009650", "frame_020850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the tote bag, the delivery bag is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002938", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "286b55a2bf", "gt_frame_id": "frame_000220", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000760", "frame_000740", "frame_002760", "frame_002800", "frame_002820", "frame_003100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the door, I am facing the bottle, and I can see the towel on the right side of the view, and the shampoo bottle near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002941", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "31a2c91c43", "gt_frame_id": "frame_000160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003240", "frame_003120", "frame_001880", "frame_001660", "frame_004770", "frame_000860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the flush button, and the bathroom mat remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002944", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "31a2c91c43", "gt_frame_id": "frame_004590", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000000", "frame_001880", "frame_002040", "frame_003120", "frame_003160", "frame_003360"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the pants, with the bathroom mat to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002947", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3864514494", "gt_frame_id": "frame_000660", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000280", "frame_000000", "frame_001060", "frame_001120", "frame_006060", "frame_006200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the serving trolley, I am facing the water boiler.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002950", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3864514494", "gt_frame_id": "frame_000580", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006220", "frame_002100", "frame_001960", "frame_001060", "frame_001100", "frame_002750"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the cup, and the serving trolley is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002953", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3864514494", "gt_frame_id": "frame_005510", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000280", "frame_001960", "frame_002040", "frame_001980", "frame_001140", "frame_000640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nMy view is toward the dishwasher, and the serving trolley is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002956", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3864514494", "gt_frame_id": "frame_005830", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000640", "frame_000580", "frame_000020", "frame_001040", "frame_000140", "frame_004850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the espresso machine, the dishwasher is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002959", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_006160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005650", "frame_005410", "frame_004370", "frame_008340", "frame_010710", "frame_009550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am positioned with the chair behind me, facing the tv.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002962", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_002210", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000890", "frame_001850", "frame_001770", "frame_001710", "frame_001630", "frame_002790"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the sink, I am facing the bag, and I can see the washing machine on the left side of the view, and the weighing scale on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002965", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_002240", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000660", "frame_000890", "frame_001850", "frame_001990", "frame_001690", "frame_002790"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the sink, I am facing the basket, and I can see the drying rack on the left side of the view, and the glass wall on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002968", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_002170", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000890", "frame_001890", "frame_001850", "frame_001830", "frame_001710", "frame_002390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the sink, I face the weighing scale, and I can see the drying rack on the left side of the view, and the window near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002971", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_005160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005310", "frame_008400", "frame_006500", "frame_004970", "frame_003870", "frame_008060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the refrigerator to my right, I am facing the bottle.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002974", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3db0a1c8f3", "gt_frame_id": "frame_001700", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003770", "frame_000910", "frame_002170", "frame_002190", "frame_002090", "frame_008160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nMy view is toward the weighing scale, and the toilet seat is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002977", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_001790", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006270", "frame_001130", "frame_007090", "frame_006950", "frame_004420", "frame_003380"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am directly between the shelf to the left and the table to the right, facing the coffee machine.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002980", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_000500", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001210", "frame_006190", "frame_004750", "frame_004930", "frame_007090", "frame_006950"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the office chair, I am facing the coffee machine, and I can see the blinds on the left side of the view, and the printer on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002983", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_001300", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004930", "frame_007090", "frame_006950", "frame_002800", "frame_003750", "frame_003260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the shelf, I face the office chair, and I can see the coffee machine on the left side of the view, and the headphone case on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002986", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_001800", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000990", "frame_001130", "frame_000520", "frame_004850", "frame_003380", "frame_000000"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the shelf, I am facing the phone, and I can see the window near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002989", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_000560", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006170", "frame_006210", "frame_000300", "frame_001280", "frame_007110", "frame_001130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the office chair to my right, I face the headphone case.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002992", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_003810", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002660", "frame_002640", "frame_006970", "frame_007090", "frame_002820", "frame_003400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the cabinet, I am facing the keyboard, and I can see the coffee machine near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002995", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "3e8bba0176", "gt_frame_id": "frame_000530", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001660", "frame_004810", "frame_006210", "frame_005210", "frame_004570", "frame_001210"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the office chair, facing the printer.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00002998", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "40aec5fffa", "gt_frame_id": "frame_005070", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001500", "frame_006660", "frame_004060", "frame_002940", "frame_002090", "frame_000020"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the trash bin, I am facing the dishwasher, and I can see the coffee maker on the right side of the view, and the sink counter near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003001", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "40aec5fffa", "gt_frame_id": "frame_001500", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004080", "frame_000650", "frame_003020", "frame_002960", "frame_004730", "frame_003220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the trash bin to my left, I face the coffee maker.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003004", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "40aec5fffa", "gt_frame_id": "frame_001630", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003220", "frame_001280", "frame_002920", "frame_000650", "frame_000370", "frame_005390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the trash bin, I face the sink counter.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003007", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "40aec5fffa", "gt_frame_id": "frame_005490", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001300", "frame_005210", "frame_003120", "frame_003060", "frame_000370", "frame_004020"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the dishwasher, I am facing the water filter, and I can see the coffee maker on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003010", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_005160", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000860", "frame_006510", "frame_004720", "frame_001520", "frame_000260", "frame_001720"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the toilet to my left, I face the bottle.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003013", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_002110", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000060", "frame_000800", "frame_000660", "frame_000600", "frame_000520", "frame_005670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the towel, with the toilet to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003016", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_004660", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000680", "frame_000840", "frame_000020", "frame_004970", "frame_006130", "frame_005710"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the towel, I am facing the trash bin, and I can see the bottle on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003019", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "45b0dac5e3", "gt_frame_id": "frame_005310", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001700", "frame_000220", "frame_000340", "frame_002420", "frame_000560", "frame_005670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the toothpaste, with the bathroom cabinet to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003022", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5748ce6f01", "gt_frame_id": "frame_001820", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004230", "frame_006100", "frame_000060", "frame_003290", "frame_003210", "frame_004850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nThe flat panel display is behind me as I face the office chair.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003025", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "578511c8a9", "gt_frame_id": "frame_013350", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_012010", "frame_011420", "frame_013100", "frame_013080", "frame_012920", "frame_011330"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the box, I am facing the keyboard.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003028", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5942004064", "gt_frame_id": "frame_009060", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003830", "frame_003870", "frame_008780", "frame_008480", "frame_003550", "frame_003410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the carpet, and the table is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003031", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5942004064", "gt_frame_id": "frame_004140", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004380", "frame_003830", "frame_003760", "frame_008760", "frame_009060", "frame_008800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the pillow, with the door to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003034", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5942004064", "gt_frame_id": "frame_008340", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008060", "frame_005720", "frame_007480", "frame_009180", "frame_008780", "frame_008760"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the plate, facing the pillow, and I can see the guitar on the left side of the view, and the table near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003037", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5eb31827b7", "gt_frame_id": "frame_007210", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006070", "frame_005570", "frame_005510", "frame_001820", "frame_004200", "frame_005170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the rolling chair to my left, I am facing the trash can.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003040", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_007600", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004910", "frame_007090", "frame_005800", "frame_006240", "frame_006200", "frame_002140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the copy machine to my left, I am facing the calendar.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003043", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5ee7c22ba0", "gt_frame_id": "frame_003840", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001420", "frame_000280", "frame_006240", "frame_006320", "frame_004420", "frame_000100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the bottle crate, the table is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003046", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "5f99900f09", "gt_frame_id": "frame_003770", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002830", "frame_002970", "frame_004390", "frame_005650", "frame_004350", "frame_004030"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the suitcase, I face the coffee maker, and I can see the dish drainer on the left side of the view, and the refrigerator on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003049", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "6115eddb86", "gt_frame_id": "frame_001720", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000400", "frame_006100", "frame_008580", "frame_005600", "frame_008260", "frame_000300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the sofa, I face the toilet, and I can see the headboard on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003052", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "6115eddb86", "gt_frame_id": "frame_008630", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000100", "frame_003570", "frame_003890", "frame_002690", "frame_002650", "frame_007890"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the sofa, with the ottoman chair to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003055", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "6115eddb86", "gt_frame_id": "frame_008620", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002470", "frame_002650", "frame_002690", "frame_001390", "frame_000120", "frame_008400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the ottoman chair, I face the standing lamp.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003058", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "6115eddb86", "gt_frame_id": "frame_000800", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008680", "frame_007850", "frame_007810", "frame_007770", "frame_001990", "frame_001910"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the chair to my right, I am facing the coffee maker.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003061", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7b6477cb95", "gt_frame_id": "frame_002610", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005850", "frame_003340", "frame_002150", "frame_001560", "frame_006890", "frame_006850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the office chair, I am facing the bookshelf, and I can see the whiteboard eraser on the left side of the view, and the storage cabinet on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003064", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7b6477cb95", "gt_frame_id": "frame_004860", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006300", "frame_006210", "frame_005740", "frame_005220", "frame_005240", "frame_005480"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the office chair, I face the keyboard, and I can see the power socket on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003067", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7b6477cb95", "gt_frame_id": "frame_007130", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002150", "frame_001440", "frame_006890", "frame_001210", "frame_001250", "frame_004990"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the keyboard, facing the telephone, with the table to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003070", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7b6477cb95", "gt_frame_id": "frame_005720", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006320", "frame_004910", "frame_004990", "frame_007110", "frame_006870", "frame_006850"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFacing the book, I am positioned between the heater to my left and the office chair to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003073", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7b6477cb95", "gt_frame_id": "frame_001650", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006320", "frame_004950", "frame_005780", "frame_006870", "frame_001230", "frame_007170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the office chair, facing the book, and I can see the mouse on the left side of the view, and the heater on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003076", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7bc286c1b6", "gt_frame_id": "frame_000390", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001420", "frame_001140", "frame_003340", "frame_000000", "frame_004110", "frame_004170"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the sink, I am facing the toilet paper rolls.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003079", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7bc286c1b6", "gt_frame_id": "frame_004180", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001300", "frame_001440", "frame_000350", "frame_000310", "frame_000450", "frame_000250"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the toothpaste, the toilet is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003082", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7bc286c1b6", "gt_frame_id": "frame_001450", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001140", "frame_003340", "frame_000370", "frame_000350", "frame_003970", "frame_003830"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am directly between the cleaning liquid to the left and the toilet to the right, facing the razor.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003085", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "7bc286c1b6", "gt_frame_id": "frame_001050", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000250", "frame_000330", "frame_003340", "frame_002780", "frame_001340", "frame_003990"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the razor, the bathroom counter is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003088", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "99fa5c25e1", "gt_frame_id": "frame_002000", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005840", "frame_005460", "frame_000310", "frame_000890", "frame_006500", "frame_006640"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the stool, I am facing the wooden brush.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003091", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_002230", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001680", "frame_001920", "frame_000180", "frame_000260", "frame_003160", "frame_003300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the sit-up pillow to my left, I am facing the speaker.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003094", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "a24f64f7fb", "gt_frame_id": "frame_002060", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000040", "frame_001780", "frame_002820", "frame_002420", "frame_000140", "frame_000320"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the sit-up pillow, facing the headphone.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003097", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "ac48a9b736", "gt_frame_id": "frame_009350", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008380", "frame_006770", "frame_007230", "frame_007450", "frame_008830", "frame_008980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the office chair to my right, I face the fire extinguisher.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003100", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_004920", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003860", "frame_002770", "frame_003960", "frame_004160", "frame_005400", "frame_010190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the chair, facing the box, and I can see the laptop on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003103", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_006990", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_010170", "frame_004560", "frame_004580", "frame_002710", "frame_003780", "frame_000550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the cabinet, I am facing the mouse, and I can see the keyboard on the left side of the view, and the laptop near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003106", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_008940", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_009490", "frame_007680", "frame_007660", "frame_005300", "frame_004000", "frame_002710"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the telephone, with the chair to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003109", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_007910", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007460", "frame_007660", "frame_005320", "frame_009240", "frame_005260", "frame_009180"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the cabinet, I face the heater, and I can see the mouse on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003112", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_010200", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000550", "frame_003050", "frame_004580", "frame_003960", "frame_004960", "frame_007560"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the cabinet, I am facing the keyboard, and I can see the laptop on the right side of the view, and the backpack near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003115", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_007940", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005260", "frame_009000", "frame_009260", "frame_008940", "frame_007640", "frame_007460"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the cabinet to my left, I face the headphone.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003118", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_009820", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004000", "frame_010190", "frame_008880", "frame_007460", "frame_007680", "frame_001690"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the cabinet, I am facing the book, and I can see the filter jug on the left side of the view, and the monitor on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003121", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "acd95847c5", "gt_frame_id": "frame_004900", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008280", "frame_005400", "frame_007620", "frame_004020", "frame_003980", "frame_003960"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the chair, facing the backpack, and I can see the laptop on the right side of the view, and the box near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003124", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "b0a08200c9", "gt_frame_id": "frame_002770", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000040", "frame_002220", "frame_008560", "frame_008580", "frame_000400", "frame_000140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am directly between the heater to the left and the storage rack to the right, facing the window.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003127", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "b0a08200c9", "gt_frame_id": "frame_002740", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_005000", "frame_008580", "frame_008560", "frame_002160", "frame_003520", "frame_003160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the box, with the heater to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003130", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "bcd2436daf", "gt_frame_id": "frame_001420", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004720", "frame_004700", "frame_004900", "frame_002620", "frame_000140", "frame_000060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the shoes, I face the plant pot, and I can see the power socket near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003133", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "bcd2436daf", "gt_frame_id": "frame_004910", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002090", "frame_000240", "frame_004460", "frame_004500", "frame_004050", "frame_000760"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the blanket to my right, I face the pillow.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003136", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "bcd2436daf", "gt_frame_id": "frame_000740", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001810", "frame_001770", "frame_004070", "frame_000100", "frame_004860", "frame_000960"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the shoes, the blanket is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003139", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "bde1e479ad", "gt_frame_id": "frame_002170", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008520", "frame_008500", "frame_007840", "frame_006770", "frame_006810", "frame_005280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the power strip, I am facing the heater.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003142", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_004970", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001840", "frame_003730", "frame_000480", "frame_003890", "frame_000640", "frame_000900"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the table, and the jacket remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003145", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_004490", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004250", "frame_002450", "frame_002370", "frame_002230", "frame_001540", "frame_001900"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am positioned between the door on the left and the pillow on the right, and I am facing the closet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003148", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_001000", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002270", "frame_002530", "frame_004030", "frame_000740", "frame_000540", "frame_003390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the mouse pad, and the nightstand is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003151", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_000940", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004890", "frame_004290", "frame_004950", "frame_000580", "frame_000640", "frame_003510"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the nightstand, I am facing the headphones, with the pillow on my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003154", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002950", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002210", "frame_002370", "frame_002590", "frame_004930", "frame_000580", "frame_001000"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the scarf, I am facing the keyboard.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003157", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_003620", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002230", "frame_001540", "frame_001320", "frame_000280", "frame_004990", "frame_003870"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nPositioned near the scarf, I am facing the bed.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003160", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002850", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001240", "frame_000280", "frame_005030", "frame_000980", "frame_004030", "frame_004270"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the headphones, with the bedside cabinet to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003163", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_003410", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000240", "frame_000480", "frame_003790", "frame_003890", "frame_000840", "frame_002590"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the scarf, I am facing the bedside cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003166", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_001040", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002230", "frame_001540", "frame_003730", "frame_000560", "frame_000660", "frame_003410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the scarf, I am facing the books.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003169", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_004310", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003730", "frame_000640", "frame_000920", "frame_002550", "frame_002410", "frame_004570"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFacing the window, I have the pillow behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003172", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002700", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003390", "frame_001020", "frame_000820", "frame_004030", "frame_004470", "frame_002410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the pliers, the bedside cabinet is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003175", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_002550", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002310", "frame_004530", "frame_004030", "frame_000740", "frame_005110", "frame_000100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFacing the shoes, I have the comforter behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003178", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_000810", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_004290", "frame_004910", "frame_000540", "frame_003830", "frame_001080", "frame_003930"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the bedframe to my right, I am facing the office chair.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003181", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_004530", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001900", "frame_001240", "frame_000400", "frame_004950", "frame_004270", "frame_002410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am positioned between the door on the left and the pillow on the right, and I am facing the curtain.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003184", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_003110", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003790", "frame_003430", "frame_000720", "frame_000960", "frame_004910", "frame_002610"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the scarf behind me, I am facing the mouse pad.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003187", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_000970", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002610", "frame_004890", "frame_004970", "frame_000720", "frame_000580", "frame_003870"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the scarf, facing the monitor, with the box to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003190", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_001070", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001540", "frame_003790", "frame_003810", "frame_000740", "frame_000560", "frame_000280"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the lamp, the scarf is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003193", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c49a8c6cff", "gt_frame_id": "frame_003940", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000560", "frame_004950", "frame_004230", "frame_000780", "frame_000940", "frame_003310"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the nightstand, I am facing the shelf, with the book on my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003196", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c50d2d1d42", "gt_frame_id": "frame_005090", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007540", "frame_002010", "frame_004160", "frame_004220", "frame_000310", "frame_008470"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the mug, with the office table to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003199", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c50d2d1d42", "gt_frame_id": "frame_008150", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001480", "frame_000290", "frame_004640", "frame_000330", "frame_006140", "frame_007560"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the headphones, I face the monitor, and I can see the box on the right side of the view, and the mousepad near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003202", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "c50d2d1d42", "gt_frame_id": "frame_005510", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007540", "frame_002010", "frame_009290", "frame_003860", "frame_000350", "frame_001060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the office chair, I face the box, and I can see the mug on the left side of the view, and the keyboard on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003205", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "cc5237fd77", "gt_frame_id": "frame_006140", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001300", "frame_004710", "frame_004470", "frame_000300", "frame_000140", "frame_005650"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom my position near the bookshelf, I face the suitcase, and I can see the chair on the right side of the view, and the books near the top of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003208", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "cc5237fd77", "gt_frame_id": "frame_008260", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_000120", "frame_007320", "frame_004710", "frame_000780", "frame_006370", "frame_000500"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the monitor, and the bookshelf is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003211", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "e7af285f7d", "gt_frame_id": "frame_007600", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007860", "frame_008270", "frame_000640", "frame_000600", "frame_000500", "frame_006600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the bath mat, I am facing the light, and I can see the cabinet near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003214", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f2dc06b1d2", "gt_frame_id": "frame_001020", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003710", "frame_000390", "frame_000330", "frame_001290", "frame_001710", "frame_001410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the office chair, I am facing the trash bin.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003217", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f2dc06b1d2", "gt_frame_id": "frame_003730", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002790", "frame_005390", "frame_003390", "frame_005710", "frame_003050", "frame_000450"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the copier to my right, I am facing the office chair.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003220", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f2dc06b1d2", "gt_frame_id": "frame_001060", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003710", "frame_000390", "frame_001430", "frame_001710", "frame_000190", "frame_000810"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the office chair, I face the box, and I can see the table on the left side of the view, and the trash bin near the bottom of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003223", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_008590", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_008310", "frame_002700", "frame_000300", "frame_000360", "frame_000780", "frame_000860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the cabinet to my left, I face the french press.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003226", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_002290", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006680", "frame_001640", "frame_001340", "frame_000280", "frame_003200", "frame_002740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWhile near the trash can, I am facing the coffee machine.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003229", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f3d64c30f8", "gt_frame_id": "frame_003220", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001840", "frame_000540", "frame_007290", "frame_007430", "frame_002760", "frame_000100"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the cabinet to my left, I face the poster.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003232", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f9f95681fd", "gt_frame_id": "frame_001760", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006320", "frame_005140", "frame_006200", "frame_000220", "frame_000960", "frame_007300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the chopping board, the kitchen towel is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003235", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f9f95681fd", "gt_frame_id": "frame_000250", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_007180", "frame_006180", "frame_006200", "frame_006240", "frame_006360", "frame_006700"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the trash bin, with the dishwasher to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003238", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f9f95681fd", "gt_frame_id": "frame_005800", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_002670", "frame_007280", "frame_001180", "frame_000240", "frame_000120", "frame_004830"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding between the shelf and the dishwasher, I am facing the kettle.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003241", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f9f95681fd", "gt_frame_id": "frame_007300", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_003010", "frame_000200", "frame_000300", "frame_001160", "frame_000760", "frame_003340"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the shelf to my left, I am facing the stove.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003244", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "f9f95681fd", "gt_frame_id": "frame_004780", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_006500", "frame_006440", "frame_000000", "frame_006200", "frame_001860", "frame_005220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am near the kitchen cabinet, facing the cup, and I can see the fire extinguisher on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003247", "dataset": "scannetpp", "task_type": "view_localization", "scene_id": "fb5a96b1a2", "gt_frame_id": "frame_005990", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["frame_001360", "frame_007650", "frame_005670", "frame_002400", "frame_007090", "frame_000840"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom a position near the sofa, I face the power socket, and the chair is on my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003250", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41069021", "gt_frame_id": "317.772", "anchor_frame_id": "308.060", "shot": 6, "contexts_frame_id": ["308.060", "309.759", "478.356", "452.967", "322.570", "328.368"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move left about 1.0 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003253", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41069021", "gt_frame_id": "318.072", "anchor_frame_id": "328.268", "shot": 6, "contexts_frame_id": ["328.268", "453.167", "478.356", "477.757", "476.357", "472.759"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003256", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41069043", "gt_frame_id": "2874.625", "anchor_frame_id": "2875.725", "shot": 6, "contexts_frame_id": ["2875.725", "2894.234", "2866.829", "2895.034", "2896.833", "2840.439"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003259", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41069046", "gt_frame_id": "2943.330", "anchor_frame_id": "2948.528", "shot": 6, "contexts_frame_id": ["2948.528", "3028.728", "2948.928", "2949.928", "2959.324", "2984.730"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003262", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41125722", "gt_frame_id": "4543.417", "anchor_frame_id": "4549.132", "shot": 6, "contexts_frame_id": ["4549.132", "4536.520", "4555.929", "4556.729", "4558.728", "4563.926"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003268", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41125722", "gt_frame_id": "4517.245", "anchor_frame_id": "4523.625", "shot": 6, "contexts_frame_id": ["4523.625", "4525.125", "4534.321", "4526.724", "4520.527", "4513.130"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move backward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003277", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41125731", "gt_frame_id": "4670.532", "anchor_frame_id": "4676.930", "shot": 6, "contexts_frame_id": ["4676.930", "4677.629", "4683.427", "4678.429", "4664.618", "4662.819"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003280", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41125731", "gt_frame_id": "4622.418", "anchor_frame_id": "4636.229", "shot": 6, "contexts_frame_id": ["4636.229", "4626.017", "4631.831", "4641.627", "4646.425", "4647.825"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003283", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41125731", "gt_frame_id": "4636.829", "anchor_frame_id": "4640.128", "shot": 6, "contexts_frame_id": ["4640.128", "4632.231", "4625.617", "4626.417", "4645.825", "4647.825"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003286", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41125756", "gt_frame_id": "5307.998", "anchor_frame_id": "5315.712", "shot": 6, "contexts_frame_id": ["5315.712", "5290.505", "5301.301", "5320.110", "5322.909", "5327.707"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move left about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003289", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159504", "gt_frame_id": "34780.577", "anchor_frame_id": "34779.677", "shot": 6, "contexts_frame_id": ["34779.677", "34738.378", "34680.785", "34692.980", "34694.979", "34697.778"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003292", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159504", "gt_frame_id": "34733.880", "anchor_frame_id": "34732.980", "shot": 6, "contexts_frame_id": ["34732.980", "34683.984", "34675.987", "34681.185", "34687.982", "34790.190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003298", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41159519", "gt_frame_id": "38288.409", "anchor_frame_id": "38280.212", "shot": 6, "contexts_frame_id": ["38280.212", "38395.398", "38394.798", "38284.211", "38247.009", "38242.411"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003304", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41159519", "gt_frame_id": "38291.108", "anchor_frame_id": "38298.804", "shot": 6, "contexts_frame_id": ["38298.804", "38300.604", "38394.398", "38394.798", "38396.397", "38396.597"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move right about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003307", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41159525", "gt_frame_id": "38090.708", "anchor_frame_id": "38081.812", "shot": 6, "contexts_frame_id": ["38081.812", "38086.110", "38094.506", "38096.905", "38098.305", "38029.117"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003310", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41159538", "gt_frame_id": "39712.335", "anchor_frame_id": "39715.034", "shot": 6, "contexts_frame_id": ["39715.034", "39723.431", "39685.630", "39688.029", "39674.234", "39671.236"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003313", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159553", "gt_frame_id": "33405.861", "anchor_frame_id": "33410.259", "shot": 6, "contexts_frame_id": ["33410.259", "33335.356", "33328.359", "33432.751", "33382.554", "33387.152"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003316", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41159553", "gt_frame_id": "33425.453", "anchor_frame_id": "33435.249", "shot": 6, "contexts_frame_id": ["33435.249", "33430.151", "33357.764", "33328.758", "33403.162", "33340.354"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003319", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159553", "gt_frame_id": "33410.060", "anchor_frame_id": "33405.861", "shot": 6, "contexts_frame_id": ["33405.861", "33452.959", "33356.964", "33335.356", "33352.965", "33344.752"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003322", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159557", "gt_frame_id": "33301.153", "anchor_frame_id": "33305.051", "shot": 6, "contexts_frame_id": ["33305.051", "33307.250", "33246.658", "33247.657", "33249.456", "33251.056"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003325", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159557", "gt_frame_id": "33244.158", "anchor_frame_id": "33245.758", "shot": 6, "contexts_frame_id": ["33245.758", "33306.251", "33309.649", "33257.653", "33240.260", "33192.262"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003328", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41159557", "gt_frame_id": "33245.158", "anchor_frame_id": "33242.959", "shot": 6, "contexts_frame_id": ["33242.959", "33240.460", "33306.251", "33309.649", "33303.252", "33192.262"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003331", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41159571", "gt_frame_id": "35044.221", "anchor_frame_id": "35015.715", "shot": 6, "contexts_frame_id": ["35015.715", "35018.014", "35022.812", "35022.412", "35027.411", "35028.010"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.0 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003334", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41159572", "gt_frame_id": "35129.120", "anchor_frame_id": "35121.523", "shot": 6, "contexts_frame_id": ["35121.523", "35120.724", "35122.323", "35094.517", "35104.713", "35105.713"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003337", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41254246", "gt_frame_id": "101393.312", "anchor_frame_id": "101391.913", "shot": 6, "contexts_frame_id": ["101391.913", "101464.199", "101433.012", "101428.597", "101427.198", "101425.598"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003340", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41254262", "gt_frame_id": "101776.403", "anchor_frame_id": "101800.293", "shot": 6, "contexts_frame_id": ["101800.293", "101884.692", "101862.501", "101861.101", "101794.496", "101810.106"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003346", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41254269", "gt_frame_id": "101717.094", "anchor_frame_id": "101705.399", "shot": 6, "contexts_frame_id": ["101705.399", "101707.798", "101741.601", "101748.398", "101747.198", "101743.200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move left about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003349", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41254382", "gt_frame_id": "94892.165", "anchor_frame_id": "94881.069", "shot": 6, "contexts_frame_id": ["94881.069", "94938.663", "94881.469", "94886.067", "94922.269", "94924.868"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003352", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41254382", "gt_frame_id": "94904.360", "anchor_frame_id": "94901.961", "shot": 6, "contexts_frame_id": ["94901.961", "94900.661", "94900.262", "94889.666", "94891.065", "94881.269"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003358", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41254412", "gt_frame_id": "95745.681", "anchor_frame_id": "95752.495", "shot": 6, "contexts_frame_id": ["95752.495", "95783.483", "95781.484", "95780.084", "95779.284", "95790.680"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003361", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "41254435", "gt_frame_id": "96264.410", "anchor_frame_id": "96263.211", "shot": 6, "contexts_frame_id": ["96263.211", "96240.403", "96240.803", "96239.803", "96236.405", "96255.214"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003364", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "41254441", "gt_frame_id": "96364.304", "anchor_frame_id": "96363.104", "shot": 6, "contexts_frame_id": ["96363.104", "96314.507", "96390.110", "96393.909", "96395.708", "96396.908"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.3 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003367", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "41254441", "gt_frame_id": "96371.201", "anchor_frame_id": "96372.717", "shot": 6, "contexts_frame_id": ["96372.717", "96333.916", "96368.102", "96376.116", "96367.503", "96378.915"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003370", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42444946", "gt_frame_id": "223685.366", "anchor_frame_id": "223687.465", "shot": 6, "contexts_frame_id": ["223687.465", "223678.468", "223682.067", "223689.864", "223690.663", "223697.677"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003373", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42444966", "gt_frame_id": "40414.680", "anchor_frame_id": "40419.978", "shot": 6, "contexts_frame_id": ["40419.978", "40431.590", "40433.589", "40441.186", "40427.991", "40425.792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003376", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42444966", "gt_frame_id": "40343.193", "anchor_frame_id": "40331.381", "shot": 6, "contexts_frame_id": ["40331.381", "40332.581", "40333.381", "40330.782", "40376.779", "40375.580"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003379", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42444968", "gt_frame_id": "40302.293", "anchor_frame_id": "40315.788", "shot": 6, "contexts_frame_id": ["40315.788", "40316.787", "40285.384", "40291.581", "40292.381", "40314.388"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003385", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42444976", "gt_frame_id": "40233.589", "anchor_frame_id": "40228.591", "shot": 6, "contexts_frame_id": ["40228.591", "40229.590", "40236.387", "40167.583", "40146.991", "40185.792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003388", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42445031", "gt_frame_id": "49454.391", "anchor_frame_id": "49470.601", "shot": 6, "contexts_frame_id": ["49470.601", "49468.402", "49472.401", "49451.192", "49478.198", "49457.390"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move left about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003391", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42445031", "gt_frame_id": "49451.492", "anchor_frame_id": "49452.592", "shot": 6, "contexts_frame_id": ["49452.592", "49445.994", "49472.601", "49471.601", "49478.198", "49455.990"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003397", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42445429", "gt_frame_id": "50358.290", "anchor_frame_id": "50352.592", "shot": 6, "contexts_frame_id": ["50352.592", "50351.092", "50347.294", "50345.495", "50344.495", "50318.089"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003403", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42445441", "gt_frame_id": "50277.588", "anchor_frame_id": "50278.588", "shot": 6, "contexts_frame_id": ["50278.588", "50256.380", "50251.382", "50247.983", "50285.385", "50286.185"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003409", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42445966", "gt_frame_id": "347103.840", "anchor_frame_id": "347098.226", "shot": 6, "contexts_frame_id": ["347098.226", "347098.826", "347111.637", "347119.234", "347162.833", "347172.429"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then move backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003412", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42445988", "gt_frame_id": "347353.038", "anchor_frame_id": "347345.824", "shot": 6, "contexts_frame_id": ["347345.824", "347348.740", "347401.535", "347426.325", "347359.935", "347382.726"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003415", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42445991", "gt_frame_id": "75278.332", "anchor_frame_id": "75268.236", "shot": 6, "contexts_frame_id": ["75268.236", "75342.822", "75265.437", "75333.626", "75292.626", "75257.024"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003418", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42445991", "gt_frame_id": "75272.435", "anchor_frame_id": "75278.032", "shot": 6, "contexts_frame_id": ["75278.032", "75259.623", "75291.627", "75256.624", "75342.822", "75344.221"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003421", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42445991", "gt_frame_id": "75335.525", "anchor_frame_id": "75336.425", "shot": 6, "contexts_frame_id": ["75336.425", "75342.822", "75277.832", "75259.623", "75257.224", "75293.026"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003424", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42445991", "gt_frame_id": "75277.532", "anchor_frame_id": "75260.423", "shot": 6, "contexts_frame_id": ["75260.423", "75335.425", "75288.428", "75259.023", "75293.626", "75295.025"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003427", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446083", "gt_frame_id": "348906.185", "anchor_frame_id": "348925.977", "shot": 6, "contexts_frame_id": ["348925.977", "348919.880", "348918.680", "348917.881", "348899.688", "348836.481"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003430", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446093", "gt_frame_id": "350166.469", "anchor_frame_id": "350178.364", "shot": 6, "contexts_frame_id": ["350178.364", "350174.065", "350172.666", "350189.859", "350190.858", "350126.868"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.0 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003436", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446100", "gt_frame_id": "77995.595", "anchor_frame_id": "77993.879", "shot": 6, "contexts_frame_id": ["77993.879", "78037.295", "78028.682", "78025.683", "78022.885", "78021.085"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003442", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446100", "gt_frame_id": "78007.391", "anchor_frame_id": "78010.389", "shot": 6, "contexts_frame_id": ["78010.389", "77953.695", "77987.682", "77988.881", "77954.495", "78002.892"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move backward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003445", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446103", "gt_frame_id": "78063.685", "anchor_frame_id": "78061.386", "shot": 6, "contexts_frame_id": ["78061.386", "78060.386", "78124.794", "78119.796", "78122.395", "78160.780"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003448", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446103", "gt_frame_id": "78107.284", "anchor_frame_id": "78100.187", "shot": 6, "contexts_frame_id": ["78100.187", "78173.392", "78094.989", "78085.993", "78084.593", "78084.194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003454", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446104", "gt_frame_id": "349854.263", "anchor_frame_id": "349832.072", "shot": 6, "contexts_frame_id": ["349832.072", "349839.969", "349889.366", "349893.564", "349858.362", "349800.569"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003457", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446104", "gt_frame_id": "349822.476", "anchor_frame_id": "349829.074", "shot": 6, "contexts_frame_id": ["349829.074", "349807.166", "349898.562", "349899.961", "349875.571", "349800.169"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003460", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446104", "gt_frame_id": "349829.373", "anchor_frame_id": "349821.877", "shot": 6, "contexts_frame_id": ["349821.877", "349892.764", "349899.961", "349875.571", "349800.169", "349801.168"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003466", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446114", "gt_frame_id": "78317.884", "anchor_frame_id": "78316.585", "shot": 6, "contexts_frame_id": ["78316.585", "78326.381", "78324.982", "78324.182", "78323.382", "78322.183"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003469", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446116", "gt_frame_id": "350027.876", "anchor_frame_id": "350048.067", "shot": 6, "contexts_frame_id": ["350048.067", "350020.062", "350012.465", "350043.869", "350058.863", "350060.062"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.6 meters, then move backward about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003472", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446116", "gt_frame_id": "350048.167", "anchor_frame_id": "350059.862", "shot": 6, "contexts_frame_id": ["350059.862", "350056.864", "350021.062", "350019.262", "350012.065", "350036.672"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003475", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446116", "gt_frame_id": "350023.161", "anchor_frame_id": "350028.375", "shot": 6, "contexts_frame_id": ["350028.375", "350059.463", "350060.662", "350075.672", "349966.068", "350019.062"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003478", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446163", "gt_frame_id": "79572.022", "anchor_frame_id": "79572.921", "shot": 6, "contexts_frame_id": ["79572.921", "79628.532", "79626.733", "79620.319", "79621.719", "79588.132"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003487", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446163", "gt_frame_id": "79629.932", "anchor_frame_id": "79614.521", "shot": 6, "contexts_frame_id": ["79614.521", "79633.930", "79605.925", "79574.121", "79576.920", "79621.719"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003490", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446527", "gt_frame_id": "198661.937", "anchor_frame_id": "198682.145", "shot": 6, "contexts_frame_id": ["198682.145", "198646.843", "198657.239", "198697.239", "198694.440", "198698.638"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move forward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003493", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446532", "gt_frame_id": "198821.038", "anchor_frame_id": "198829.634", "shot": 6, "contexts_frame_id": ["198829.634", "198909.735", "198823.537", "198902.938", "198847.344", "198866.136"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003499", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446540", "gt_frame_id": "201521.640", "anchor_frame_id": "201523.040", "shot": 6, "contexts_frame_id": ["201523.040", "201475.543", "201477.942", "201480.341", "201534.135", "201495.951"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003502", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42446540", "gt_frame_id": "201513.844", "anchor_frame_id": "201510.745", "shot": 6, "contexts_frame_id": ["201510.745", "201478.142", "201534.535", "201495.951", "201497.750", "201536.151"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003505", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446540", "gt_frame_id": "201500.849", "anchor_frame_id": "201501.848", "shot": 6, "contexts_frame_id": ["201501.848", "201506.147", "201509.146", "201477.542", "201473.144", "201469.745"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003508", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42446541", "gt_frame_id": "201667.747", "anchor_frame_id": "201670.246", "shot": 6, "contexts_frame_id": ["201670.246", "201700.833", "201660.233", "201682.241", "201683.041", "201684.440"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003511", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42897501", "gt_frame_id": "270049.586", "anchor_frame_id": "270059.982", "shot": 6, "contexts_frame_id": ["270059.982", "270058.982", "270038.191", "270041.989", "270045.188", "270035.792"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003514", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897504", "gt_frame_id": "269862.479", "anchor_frame_id": "269872.192", "shot": 6, "contexts_frame_id": ["269872.192", "269878.889", "269867.894", "269874.891", "269844.486", "269844.687"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003517", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42897521", "gt_frame_id": "273959.095", "anchor_frame_id": "273959.995", "shot": 6, "contexts_frame_id": ["273959.995", "274010.091", "273973.306", "274008.492", "274004.293", "273978.304"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003520", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897526", "gt_frame_id": "273856.704", "anchor_frame_id": "273846.791", "shot": 6, "contexts_frame_id": ["273846.791", "273859.802", "273852.606", "273851.406", "273879.195", "273878.795"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003523", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897541", "gt_frame_id": "275386.231", "anchor_frame_id": "275388.930", "shot": 6, "contexts_frame_id": ["275388.930", "275436.127", "275395.127", "275428.730", "275424.932", "275443.141"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.9 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003532", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897541", "gt_frame_id": "275386.231", "anchor_frame_id": "275378.534", "shot": 6, "contexts_frame_id": ["275378.534", "275379.734", "275436.127", "275395.127", "275430.530", "275427.531"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003535", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897541", "gt_frame_id": "275389.230", "anchor_frame_id": "275386.231", "shot": 6, "contexts_frame_id": ["275386.231", "275435.528", "275432.329", "275395.127", "275429.730", "275427.131"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003538", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897542", "gt_frame_id": "275456.436", "anchor_frame_id": "275476.427", "shot": 6, "contexts_frame_id": ["275476.427", "275477.127", "275479.126", "275479.726", "275460.734", "275481.325"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003541", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897561", "gt_frame_id": "278950.008", "anchor_frame_id": "278948.709", "shot": 6, "contexts_frame_id": ["278948.709", "278924.302", "278926.701", "278929.717", "278942.312", "278942.711"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003544", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897599", "gt_frame_id": "476363.204", "anchor_frame_id": "476357.006", "shot": 6, "contexts_frame_id": ["476357.006", "476511.610", "476564.005", "476402.804", "476562.406", "476413.400"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003547", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42897599", "gt_frame_id": "476527.703", "anchor_frame_id": "476535.000", "shot": 6, "contexts_frame_id": ["476535.000", "476533.001", "476493.600", "476458.598", "476461.397", "476501.997"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003550", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897599", "gt_frame_id": "476425.711", "anchor_frame_id": "476442.105", "shot": 6, "contexts_frame_id": ["476442.105", "476337.414", "476435.208", "476517.207", "476466.811", "476563.005"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move right about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003556", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897629", "gt_frame_id": "477095.703", "anchor_frame_id": "477077.994", "shot": 6, "contexts_frame_id": ["477077.994", "477078.294", "477078.693", "477083.891", "477191.997", "477190.597"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003559", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897629", "gt_frame_id": "477095.503", "anchor_frame_id": "477115.795", "shot": 6, "contexts_frame_id": ["477115.795", "477078.294", "477083.691", "477084.091", "477192.197", "477190.597"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.9 meters, then move right about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003562", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897647", "gt_frame_id": "478218.188", "anchor_frame_id": "478209.875", "shot": 6, "contexts_frame_id": ["478209.875", "478207.576", "478225.385", "478230.183", "478232.183", "478233.182"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003565", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897647", "gt_frame_id": "478213.174", "anchor_frame_id": "478205.677", "shot": 6, "contexts_frame_id": ["478205.677", "478207.976", "478209.975", "478203.378", "478198.580", "478192.382"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.1 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003577", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897651", "gt_frame_id": "478449.277", "anchor_frame_id": "478431.084", "shot": 6, "contexts_frame_id": ["478431.084", "478452.675", "478428.085", "478436.282", "478439.681", "478440.680"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003580", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897651", "gt_frame_id": "478392.383", "anchor_frame_id": "478389.984", "shot": 6, "contexts_frame_id": ["478389.984", "478343.087", "478425.286", "478426.886", "478431.484", "478449.077"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003586", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897667", "gt_frame_id": "478036.180", "anchor_frame_id": "478029.482", "shot": 6, "contexts_frame_id": ["478029.482", "478006.475", "478027.283", "478033.681", "478038.878", "478119.479"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003589", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897667", "gt_frame_id": "478024.684", "anchor_frame_id": "478040.478", "shot": 6, "contexts_frame_id": ["478040.478", "478119.079", "478034.680", "478028.083", "478005.675", "478004.876"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move left about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003595", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897688", "gt_frame_id": "479187.280", "anchor_frame_id": "479188.179", "shot": 6, "contexts_frame_id": ["479188.179", "479181.082", "479207.688", "479199.092", "479192.078", "479313.078"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003601", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897688", "gt_frame_id": "479187.380", "anchor_frame_id": "479176.384", "shot": 6, "contexts_frame_id": ["479176.384", "479182.082", "479181.082", "479199.092", "479192.078", "479313.078"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.2 meters, then move backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003607", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42897688", "gt_frame_id": "479192.278", "anchor_frame_id": "479199.391", "shot": 6, "contexts_frame_id": ["479199.391", "479187.080", "479152.477", "479151.678", "479169.687", "479310.679"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.5 meters, then move backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003610", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897688", "gt_frame_id": "479258.684", "anchor_frame_id": "479249.987", "shot": 6, "contexts_frame_id": ["479249.987", "479251.087", "479256.085", "479262.682", "479236.076", "479181.082"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003613", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42897688", "gt_frame_id": "479203.690", "anchor_frame_id": "479223.481", "shot": 6, "contexts_frame_id": ["479223.481", "479188.279", "479151.878", "479192.278", "479193.477", "479197.476"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003619", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897695", "gt_frame_id": "480362.899", "anchor_frame_id": "480341.791", "shot": 6, "contexts_frame_id": ["480341.791", "480275.002", "480281.000", "480376.794", "480241.599", "480381.392"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then turn around.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003622", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42897695", "gt_frame_id": "480363.299", "anchor_frame_id": "480356.402", "shot": 6, "contexts_frame_id": ["480356.402", "480376.993", "480241.199", "480381.592", "480239.400", "480245.398"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003625", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897696", "gt_frame_id": "480560.701", "anchor_frame_id": "480575.195", "shot": 6, "contexts_frame_id": ["480575.195", "480574.295", "480527.498", "480545.890", "480540.693", "480537.694"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003628", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42897696", "gt_frame_id": "480575.295", "anchor_frame_id": "480547.990", "shot": 6, "contexts_frame_id": ["480547.990", "480520.901", "480527.498", "480527.298", "480561.101", "480587.290"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003631", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898519", "gt_frame_id": "750502.167", "anchor_frame_id": "750507.365", "shot": 6, "contexts_frame_id": ["750507.365", "750493.771", "750494.371", "750505.966", "750510.164", "750518.178"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003634", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898521", "gt_frame_id": "750386.265", "anchor_frame_id": "750381.967", "shot": 6, "contexts_frame_id": ["750381.967", "750381.567", "750383.566", "750392.179", "750379.568", "750367.772"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003643", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898527", "gt_frame_id": "750718.262", "anchor_frame_id": "750729.074", "shot": 6, "contexts_frame_id": ["750729.074", "750724.276", "750707.067", "750736.271", "750732.273", "750754.664"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003646", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898527", "gt_frame_id": "750751.565", "anchor_frame_id": "750759.762", "shot": 6, "contexts_frame_id": ["750759.762", "750756.063", "750736.271", "750734.472", "750727.875", "750718.462"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees, then move right about 1.2 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003649", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898527", "gt_frame_id": "750655.571", "anchor_frame_id": "750660.469", "shot": 6, "contexts_frame_id": ["750660.469", "750648.474", "750621.069", "750621.669", "750666.467", "750678.862"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003652", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898538", "gt_frame_id": "752004.554", "anchor_frame_id": "751978.248", "shot": 6, "contexts_frame_id": ["751978.248", "752058.948", "752055.349", "751998.956", "751960.355", "751957.756"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move left about 1.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003655", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898538", "gt_frame_id": "751999.756", "anchor_frame_id": "752011.751", "shot": 6, "contexts_frame_id": ["752011.751", "752004.154", "752002.754", "752055.949", "752057.748", "751976.149"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.5 meters, then move right about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003658", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898544", "gt_frame_id": "752133.850", "anchor_frame_id": "752137.149", "shot": 6, "contexts_frame_id": ["752137.149", "752141.747", "752129.752", "752130.152", "752123.954", "752123.754"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003661", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898570", "gt_frame_id": "752862.278", "anchor_frame_id": "752883.786", "shot": 6, "contexts_frame_id": ["752883.786", "752872.974", "752871.375", "752825.377", "752830.375", "752854.582"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003664", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898570", "gt_frame_id": "752866.277", "anchor_frame_id": "752844.586", "shot": 6, "contexts_frame_id": ["752844.586", "752830.375", "752825.177", "752871.375", "752874.373", "752878.372"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.4 meters, then move right about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003673", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898570", "gt_frame_id": "752878.272", "anchor_frame_id": "752899.980", "shot": 6, "contexts_frame_id": ["752899.980", "752904.378", "752891.983", "752889.984", "752883.386", "752872.974"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees, then move backward about 1.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003676", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898570", "gt_frame_id": "752884.086", "anchor_frame_id": "752866.277", "shot": 6, "contexts_frame_id": ["752866.277", "752867.376", "752878.572", "752892.383", "752900.180", "752904.378"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 2.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003685", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898581", "gt_frame_id": "753256.983", "anchor_frame_id": "753285.471", "shot": 6, "contexts_frame_id": ["753285.471", "753253.367", "753070.976", "753080.572", "753359.973", "753087.969"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003691", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898811", "gt_frame_id": "167480.400", "anchor_frame_id": "167475.802", "shot": 6, "contexts_frame_id": ["167475.802", "167477.401", "167511.587", "167501.791", "167552.586", "167553.786"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003697", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898816", "gt_frame_id": "167394.102", "anchor_frame_id": "167393.102", "shot": 6, "contexts_frame_id": ["167393.102", "167367.097", "167359.100", "167362.498", "167403.298", "167401.499"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003700", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898816", "gt_frame_id": "167393.202", "anchor_frame_id": "167394.102", "shot": 6, "contexts_frame_id": ["167394.102", "167401.899", "167386.688", "167364.298", "167367.097", "167377.092"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003703", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898817", "gt_frame_id": "167416.793", "anchor_frame_id": "167413.494", "shot": 6, "contexts_frame_id": ["167413.494", "167419.492", "167420.491", "167414.094", "167453.294", "167452.694"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003706", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898817", "gt_frame_id": "167439.800", "anchor_frame_id": "167435.901", "shot": 6, "contexts_frame_id": ["167435.901", "167418.892", "167417.292", "167415.893", "167413.494", "167421.291"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003709", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898818", "gt_frame_id": "168163.176", "anchor_frame_id": "168168.074", "shot": 6, "contexts_frame_id": ["168168.074", "168171.772", "168169.373", "168167.374", "168195.579", "168188.782"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003712", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898849", "gt_frame_id": "196473.924", "anchor_frame_id": "196477.223", "shot": 6, "contexts_frame_id": ["196477.223", "196478.522", "196544.913", "196540.114", "196547.128", "196547.928"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003715", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42898854", "gt_frame_id": "196186.121", "anchor_frame_id": "196188.520", "shot": 6, "contexts_frame_id": ["196188.520", "196226.522", "196189.720", "196295.711", "196259.525", "196262.924"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003718", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898862", "gt_frame_id": "196413.115", "anchor_frame_id": "196411.615", "shot": 6, "contexts_frame_id": ["196411.615", "196405.718", "196402.119", "196419.712", "196400.920", "196389.724"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003721", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898862", "gt_frame_id": "196411.615", "anchor_frame_id": "196413.015", "shot": 6, "contexts_frame_id": ["196413.015", "196420.712", "196424.527", "196405.918", "196406.917", "196408.117"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003724", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898867", "gt_frame_id": "197761.324", "anchor_frame_id": "197757.726", "shot": 6, "contexts_frame_id": ["197757.726", "197718.325", "197725.938", "197755.327", "197752.528", "197749.929"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003727", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42898867", "gt_frame_id": "197757.726", "anchor_frame_id": "197761.324", "shot": 6, "contexts_frame_id": ["197761.324", "197717.125", "197719.124", "197727.138", "197727.938", "197729.937"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003730", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898868", "gt_frame_id": "197691.135", "anchor_frame_id": "197662.230", "shot": 6, "contexts_frame_id": ["197662.230", "197664.829", "197668.028", "197660.031", "197657.832", "197656.233"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003733", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42898869", "gt_frame_id": "197831.730", "anchor_frame_id": "197837.727", "shot": 6, "contexts_frame_id": ["197837.727", "197797.826", "197799.226", "197836.028", "197793.228", "197791.229"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move backward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003736", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42899461", "gt_frame_id": "195815.223", "anchor_frame_id": "195816.822", "shot": 6, "contexts_frame_id": ["195816.822", "195836.431", "195859.222", "195877.031", "195845.827", "195862.020"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003739", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42899471", "gt_frame_id": "195722.228", "anchor_frame_id": "195720.029", "shot": 6, "contexts_frame_id": ["195720.029", "195700.321", "195715.731", "195713.932", "195712.932", "195739.521"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003745", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42899611", "gt_frame_id": "228403.815", "anchor_frame_id": "228404.715", "shot": 6, "contexts_frame_id": ["228404.715", "228401.316", "228394.119", "228407.114", "228407.713", "228408.513"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003748", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "42899611", "gt_frame_id": "228377.625", "anchor_frame_id": "228371.028", "shot": 6, "contexts_frame_id": ["228371.028", "228373.927", "228370.511", "228411.112", "228387.122", "228405.114"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003751", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42899612", "gt_frame_id": "228499.827", "anchor_frame_id": "228491.314", "shot": 6, "contexts_frame_id": ["228491.314", "228490.914", "228491.713", "228503.925", "228492.913", "228494.312"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003754", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42899619", "gt_frame_id": "228901.568", "anchor_frame_id": "228890.372", "shot": 6, "contexts_frame_id": ["228890.372", "228889.273", "228895.270", "228906.066", "228906.466", "228906.866"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003757", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "42899620", "gt_frame_id": "228854.570", "anchor_frame_id": "228857.369", "shot": 6, "contexts_frame_id": ["228857.369", "228849.672", "228858.868", "228862.067", "228871.663", "228872.263"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003763", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42899736", "gt_frame_id": "801433.037", "anchor_frame_id": "801429.439", "shot": 6, "contexts_frame_id": ["801429.439", "801378.943", "801375.145", "801379.143", "801379.743", "801381.142"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003766", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "42899739", "gt_frame_id": "801481.534", "anchor_frame_id": "801474.037", "shot": 6, "contexts_frame_id": ["801474.037", "801512.738", "801471.938", "801516.536", "801517.336", "801461.342"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 1.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003769", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "44358435", "gt_frame_id": "42575.686", "anchor_frame_id": "42573.687", "shot": 6, "contexts_frame_id": ["42573.687", "42482.791", "42508.597", "42482.991", "42546.798", "42532.187"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003772", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358446", "gt_frame_id": "43020.097", "anchor_frame_id": "43016.798", "shot": 6, "contexts_frame_id": ["43016.798", "43029.293", "43014.899", "43012.500", "43048.302", "43051.300"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003775", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "44358448", "gt_frame_id": "42988.893", "anchor_frame_id": "42986.994", "shot": 6, "contexts_frame_id": ["42986.994", "42971.400", "42970.001", "42997.589", "42964.403", "42953.791"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003778", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358448", "gt_frame_id": "42966.102", "anchor_frame_id": "42996.090", "shot": 6, "contexts_frame_id": ["42996.090", "42996.790", "42998.589", "42969.201", "42963.004", "42952.791"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003781", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358448", "gt_frame_id": "42959.089", "anchor_frame_id": "42953.491", "shot": 6, "contexts_frame_id": ["42953.491", "42955.990", "42993.791", "42964.403", "42989.393", "42988.193"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003784", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358455", "gt_frame_id": "43407.890", "anchor_frame_id": "43396.295", "shot": 6, "contexts_frame_id": ["43396.295", "43481.493", "43481.093", "43411.489", "43479.294", "43422.101"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003787", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "44358455", "gt_frame_id": "43465.100", "anchor_frame_id": "43462.284", "shot": 6, "contexts_frame_id": ["43462.284", "43461.085", "43459.685", "43450.489", "43445.691", "43443.492"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003790", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358498", "gt_frame_id": "50516.693", "anchor_frame_id": "50515.694", "shot": 6, "contexts_frame_id": ["50515.694", "50506.597", "50507.597", "50473.194", "50474.194", "50475.194"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003793", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358498", "gt_frame_id": "50482.607", "anchor_frame_id": "50509.496", "shot": 6, "contexts_frame_id": ["50509.496", "50475.194", "50508.397", "50507.597", "50473.194", "50471.795"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003799", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "44358499", "gt_frame_id": "50567.106", "anchor_frame_id": "50565.706", "shot": 6, "contexts_frame_id": ["50565.706", "50605.590", "50608.405", "50606.006", "50575.402", "50576.202"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003802", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "44358505", "gt_frame_id": "50758.093", "anchor_frame_id": "50759.692", "shot": 6, "contexts_frame_id": ["50759.692", "50738.101", "50738.501", "50770.688", "50736.302", "50763.291"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003805", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358513", "gt_frame_id": "52684.805", "anchor_frame_id": "52657.417", "shot": 6, "contexts_frame_id": ["52657.417", "52678.708", "52654.518", "52655.118", "52651.702", "52649.903"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.4 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003808", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358513", "gt_frame_id": "52685.205", "anchor_frame_id": "52672.810", "shot": 6, "contexts_frame_id": ["52672.810", "52660.715", "52678.108", "52657.117", "52654.318", "52652.519"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move forward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003814", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358515", "gt_frame_id": "52609.603", "anchor_frame_id": "52606.604", "shot": 6, "contexts_frame_id": ["52606.604", "52626.313", "52627.712", "52601.107", "52599.307", "52596.309"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move forward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003820", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358518", "gt_frame_id": "52899.799", "anchor_frame_id": "52903.314", "shot": 6, "contexts_frame_id": ["52903.314", "52718.608", "52716.409", "52735.601", "52735.001", "52733.602"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then move backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003823", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358518", "gt_frame_id": "52903.514", "anchor_frame_id": "52895.901", "shot": 6, "contexts_frame_id": ["52895.901", "52867.013", "52710.011", "52719.207", "52772.602", "52751.611"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.5 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003826", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358532", "gt_frame_id": "53591.479", "anchor_frame_id": "53588.264", "shot": 6, "contexts_frame_id": ["53588.264", "53615.669", "53608.672", "53677.277", "53603.674", "53606.473"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003829", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358536", "gt_frame_id": "53519.575", "anchor_frame_id": "53524.174", "shot": 6, "contexts_frame_id": ["53524.174", "53522.474", "53541.266", "53540.667", "53517.093", "53538.068"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move backward about 1.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003832", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358538", "gt_frame_id": "53579.767", "anchor_frame_id": "53555.177", "shot": 6, "contexts_frame_id": ["53555.177", "53553.878", "53554.078", "53555.477", "53559.875", "53561.075"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003835", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358538", "gt_frame_id": "53554.078", "anchor_frame_id": "53579.767", "shot": 6, "contexts_frame_id": ["53579.767", "53557.276", "53556.877", "53559.076", "53559.875", "53562.674"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 2.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003838", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "44358584", "gt_frame_id": "55650.926", "anchor_frame_id": "55646.728", "shot": 6, "contexts_frame_id": ["55646.728", "55648.227", "55685.229", "55653.425", "55660.023", "55642.030"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.9 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003841", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "44358584", "gt_frame_id": "55665.920", "anchor_frame_id": "55681.831", "shot": 6, "contexts_frame_id": ["55681.831", "55683.430", "55652.426", "55670.419", "55649.827", "55686.629"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003850", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260856", "gt_frame_id": "638.592", "anchor_frame_id": "633.778", "shot": 6, "contexts_frame_id": ["633.778", "635.477", "685.690", "684.490", "660.083", "676.876"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003853", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260856", "gt_frame_id": "637.376", "anchor_frame_id": "638.193", "shot": 6, "contexts_frame_id": ["638.193", "685.690", "724.690", "725.290", "709.280", "699.684"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003856", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45260856", "gt_frame_id": "674.378", "anchor_frame_id": "687.889", "shot": 6, "contexts_frame_id": ["687.889", "686.089", "663.482", "667.081", "713.278", "640.891"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003859", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260857", "gt_frame_id": "522.990", "anchor_frame_id": "523.890", "shot": 6, "contexts_frame_id": ["523.890", "511.478", "532.686", "530.687", "541.483", "509.279"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003868", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260903", "gt_frame_id": "1583.688", "anchor_frame_id": "1585.487", "shot": 6, "contexts_frame_id": ["1585.487", "1548.302", "1571.293", "1570.693", "1569.694", "1524.095"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003871", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260903", "gt_frame_id": "1533.591", "anchor_frame_id": "1527.994", "shot": 6, "contexts_frame_id": ["1527.994", "1529.693", "1529.893", "1569.494", "1570.693", "1547.503"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003874", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260903", "gt_frame_id": "1536.390", "anchor_frame_id": "1533.191", "shot": 6, "contexts_frame_id": ["1533.191", "1530.892", "1569.494", "1570.693", "1547.503", "1548.302"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.3 meters, then move right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003877", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260903", "gt_frame_id": "1595.500", "anchor_frame_id": "1593.201", "shot": 6, "contexts_frame_id": ["1593.201", "1543.887", "1544.087", "1572.693", "1559.898", "1560.697"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003880", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45260903", "gt_frame_id": "1569.494", "anchor_frame_id": "1570.593", "shot": 6, "contexts_frame_id": ["1570.593", "1521.096", "1532.692", "1530.892", "1524.095", "1527.094"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003883", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260905", "gt_frame_id": "1460.187", "anchor_frame_id": "1483.195", "shot": 6, "contexts_frame_id": ["1483.195", "1507.202", "1506.202", "1505.003", "1474.998", "1475.598"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003889", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45260920", "gt_frame_id": "2541.223", "anchor_frame_id": "2529.627", "shot": 6, "contexts_frame_id": ["2529.627", "2484.828", "2483.629", "2480.430", "2514.017", "2478.431"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003895", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45260925", "gt_frame_id": "2427.818", "anchor_frame_id": "2438.430", "shot": 6, "contexts_frame_id": ["2438.430", "2399.829", "2404.827", "2433.015", "2453.224", "2451.225"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003898", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45260925", "gt_frame_id": "2438.430", "anchor_frame_id": "2427.818", "shot": 6, "contexts_frame_id": ["2427.818", "2406.026", "2401.628", "2434.432", "2395.230", "2445.827"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003901", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45260925", "gt_frame_id": "2442.528", "anchor_frame_id": "2422.820", "shot": 6, "contexts_frame_id": ["2422.820", "2408.225", "2420.620", "2409.025", "2426.818", "2433.015"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003904", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45260925", "gt_frame_id": "2439.130", "anchor_frame_id": "2425.818", "shot": 6, "contexts_frame_id": ["2425.818", "2433.015", "2399.829", "2400.628", "2415.423", "2445.827"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003907", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260928", "gt_frame_id": "2344.817", "anchor_frame_id": "2346.716", "shot": 6, "contexts_frame_id": ["2346.716", "2372.623", "2371.023", "2369.224", "2365.825", "2361.827"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003910", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45260928", "gt_frame_id": "2377.621", "anchor_frame_id": "2360.927", "shot": 6, "contexts_frame_id": ["2360.927", "2350.215", "2361.827", "2367.225", "2369.224", "2346.416"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003913", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45260928", "gt_frame_id": "2374.422", "anchor_frame_id": "2348.516", "shot": 6, "contexts_frame_id": ["2348.516", "2347.016", "2345.017", "2340.219", "2386.217", "2356.229"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 1.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003919", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45260928", "gt_frame_id": "2348.416", "anchor_frame_id": "2374.422", "shot": 6, "contexts_frame_id": ["2374.422", "2345.817", "2369.624", "2367.225", "2364.426", "2378.820"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.4 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003922", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45260928", "gt_frame_id": "2376.921", "anchor_frame_id": "2375.622", "shot": 6, "contexts_frame_id": ["2375.622", "2348.216", "2369.624", "2367.225", "2361.827", "2350.215"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003925", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261121", "gt_frame_id": "3463.856", "anchor_frame_id": "3437.667", "shot": 6, "contexts_frame_id": ["3437.667", "3438.467", "3443.664", "3456.459", "3446.863", "3455.060"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003928", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261121", "gt_frame_id": "3437.567", "anchor_frame_id": "3463.956", "shot": 6, "contexts_frame_id": ["3463.956", "3466.455", "3422.456", "3457.059", "3475.268", "3474.668"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003931", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261121", "gt_frame_id": "3420.057", "anchor_frame_id": "3415.459", "shot": 6, "contexts_frame_id": ["3415.459", "3457.059", "3444.064", "3441.265", "3472.669", "3461.457"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003937", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261128", "gt_frame_id": "3532.861", "anchor_frame_id": "3527.263", "shot": 6, "contexts_frame_id": ["3527.263", "3550.154", "3525.764", "3536.759", "3537.359", "3560.566"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.4 meters, then move left about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003943", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261128", "gt_frame_id": "3533.561", "anchor_frame_id": "3525.864", "shot": 6, "contexts_frame_id": ["3525.864", "3525.164", "3526.164", "3528.163", "3517.967", "3548.954"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003946", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45261129", "gt_frame_id": "3356.867", "anchor_frame_id": "3362.065", "shot": 6, "contexts_frame_id": ["3362.065", "3367.563", "3368.962", "3370.561", "3397.367", "3401.565"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003949", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261129", "gt_frame_id": "3381.057", "anchor_frame_id": "3371.661", "shot": 6, "contexts_frame_id": ["3371.661", "3359.366", "3391.169", "3360.366", "3361.565", "3365.763"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move left about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003952", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261129", "gt_frame_id": "3372.261", "anchor_frame_id": "3353.468", "shot": 6, "contexts_frame_id": ["3353.468", "3352.569", "3353.968", "3386.771", "3381.757", "3358.966"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move forward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003955", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261129", "gt_frame_id": "3357.867", "anchor_frame_id": "3381.457", "shot": 6, "contexts_frame_id": ["3381.457", "3388.171", "3381.757", "3386.571", "3397.567", "3397.967"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.6 meters, then move backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003961", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261129", "gt_frame_id": "3393.069", "anchor_frame_id": "3395.968", "shot": 6, "contexts_frame_id": ["3395.968", "3359.366", "3381.357", "3360.566", "3361.165", "3405.763"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003964", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261133", "gt_frame_id": "4038.169", "anchor_frame_id": "4036.370", "shot": 6, "contexts_frame_id": ["4036.370", "4105.258", "4102.659", "4091.064", "4009.864", "4010.664"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.6 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003967", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261140", "gt_frame_id": "3981.059", "anchor_frame_id": "3969.664", "shot": 6, "contexts_frame_id": ["3969.664", "3972.263", "3984.674", "3973.262", "3974.662", "3976.061"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.5 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003970", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261140", "gt_frame_id": "3953.771", "anchor_frame_id": "3956.270", "shot": 6, "contexts_frame_id": ["3956.270", "3951.272", "3950.472", "3962.267", "3963.666", "3967.265"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 1.1 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003973", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261142", "gt_frame_id": "3926.465", "anchor_frame_id": "3913.071", "shot": 6, "contexts_frame_id": ["3913.071", "3910.472", "3909.872", "3906.673", "3905.874", "3886.865"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003976", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261142", "gt_frame_id": "3899.760", "anchor_frame_id": "3928.764", "shot": 6, "contexts_frame_id": ["3928.764", "3929.864", "3927.865", "3906.673", "3905.874", "3886.865"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003979", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261144", "gt_frame_id": "4714.701", "anchor_frame_id": "4706.904", "shot": 6, "contexts_frame_id": ["4706.904", "4706.404", "4679.798", "4680.798", "4683.197", "4717.799"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003982", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45261179", "gt_frame_id": "2878.032", "anchor_frame_id": "2881.830", "shot": 6, "contexts_frame_id": ["2881.830", "2883.230", "2872.834", "2873.634", "2905.238", "2903.639"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003985", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261185", "gt_frame_id": "3599.182", "anchor_frame_id": "3573.592", "shot": 6, "contexts_frame_id": ["3573.592", "3585.887", "3589.086", "3592.285", "3551.884", "3608.878"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003988", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261193", "gt_frame_id": "3637.983", "anchor_frame_id": "3638.883", "shot": 6, "contexts_frame_id": ["3638.883", "3631.586", "3671.787", "3629.587", "3656.193", "3645.980"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003991", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261581", "gt_frame_id": "77546.511", "anchor_frame_id": "77531.318", "shot": 6, "contexts_frame_id": ["77531.318", "77600.505", "77537.115", "77534.916", "77550.510", "77533.117"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003994", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261619", "gt_frame_id": "55073.783", "anchor_frame_id": "55067.085", "shot": 6, "contexts_frame_id": ["55067.085", "55127.878", "55128.494", "55077.881", "55101.088", "55114.683"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00003997", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261619", "gt_frame_id": "55130.294", "anchor_frame_id": "55128.178", "shot": 6, "contexts_frame_id": ["55128.178", "55077.481", "55078.881", "55072.883", "55101.288", "55113.084"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004009", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261620", "gt_frame_id": "54941.085", "anchor_frame_id": "54943.984", "shot": 6, "contexts_frame_id": ["54943.984", "54988.483", "54960.277", "54979.486", "54967.891", "54968.291"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004012", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45261620", "gt_frame_id": "54979.786", "anchor_frame_id": "54966.292", "shot": 6, "contexts_frame_id": ["54966.292", "54941.685", "54960.277", "54958.278", "54957.679", "54968.091"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters, then move backward about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004015", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45261632", "gt_frame_id": "55915.699", "anchor_frame_id": "55904.804", "shot": 6, "contexts_frame_id": ["55904.804", "55899.189", "55933.392", "55933.792", "55944.588", "55919.398"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around, then move backward about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004018", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45662921", "gt_frame_id": "21856.616", "anchor_frame_id": "21858.215", "shot": 6, "contexts_frame_id": ["21858.215", "21850.118", "21784.929", "21776.316", "21844.921", "21843.921"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.3 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004021", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662924", "gt_frame_id": "21590.726", "anchor_frame_id": "21599.323", "shot": 6, "contexts_frame_id": ["21599.323", "21602.921", "21599.922", "21576.932", "21587.128", "21611.318"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004024", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45662924", "gt_frame_id": "21590.826", "anchor_frame_id": "21615.216", "shot": 6, "contexts_frame_id": ["21615.216", "21616.316", "21598.923", "21602.921", "21599.922", "21587.328"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.7 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004033", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662924", "gt_frame_id": "21562.221", "anchor_frame_id": "21563.921", "shot": 6, "contexts_frame_id": ["21563.921", "21558.323", "21564.720", "21575.516", "21576.932", "21587.727"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004036", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662924", "gt_frame_id": "21582.829", "anchor_frame_id": "21580.330", "shot": 6, "contexts_frame_id": ["21580.330", "21610.318", "21587.727", "21588.927", "21603.921", "21598.723"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004039", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662926", "gt_frame_id": "21719.623", "anchor_frame_id": "21722.621", "shot": 6, "contexts_frame_id": ["21722.621", "21729.319", "21659.131", "21690.718", "21687.519", "21714.525"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004042", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45662926", "gt_frame_id": "21650.118", "anchor_frame_id": "21651.418", "shot": 6, "contexts_frame_id": ["21651.418", "21660.331", "21667.928", "21705.729", "21721.722", "21743.329"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004048", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662943", "gt_frame_id": "23100.568", "anchor_frame_id": "23106.465", "shot": 6, "contexts_frame_id": ["23106.465", "23032.263", "23031.863", "23031.663", "23031.463", "23110.064"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters, then move backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004051", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45662944", "gt_frame_id": "22981.267", "anchor_frame_id": "22974.970", "shot": 6, "contexts_frame_id": ["22974.970", "22974.170", "23007.356", "22954.961", "22956.561", "22956.161"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004054", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662970", "gt_frame_id": "26113.033", "anchor_frame_id": "26107.935", "shot": 6, "contexts_frame_id": ["26107.935", "26117.531", "26119.330", "26110.334", "26110.134", "26123.728"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004057", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662975", "gt_frame_id": "26069.433", "anchor_frame_id": "26067.134", "shot": 6, "contexts_frame_id": ["26067.134", "26066.534", "26065.535", "26085.727", "26088.726", "26074.331"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004060", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662981", "gt_frame_id": "28010.487", "anchor_frame_id": "28007.389", "shot": 6, "contexts_frame_id": ["28007.389", "27942.098", "28005.689", "27967.088", "27993.494", "27982.898"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters, then move left about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004072", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45662983", "gt_frame_id": "27827.793", "anchor_frame_id": "27824.294", "shot": 6, "contexts_frame_id": ["27824.294", "27822.295", "27855.898", "27820.896", "27800.287", "27799.287"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004075", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45662983", "gt_frame_id": "27852.183", "anchor_frame_id": "27848.085", "shot": 6, "contexts_frame_id": ["27848.085", "27805.885", "27806.285", "27809.683", "27810.683", "27812.082"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004078", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45662983", "gt_frame_id": "27848.185", "anchor_frame_id": "27852.083", "shot": 6, "contexts_frame_id": ["27852.083", "27808.684", "27803.886", "27811.483", "27814.298", "27815.498"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004081", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "45662987", "gt_frame_id": "27907.695", "anchor_frame_id": "27904.696", "shot": 6, "contexts_frame_id": ["27904.696", "27901.797", "27921.189", "27880.988", "27882.188", "27912.793"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004084", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663099", "gt_frame_id": "54950.855", "anchor_frame_id": "54932.745", "shot": 6, "contexts_frame_id": ["54932.745", "54908.056", "54897.243", "54895.244", "54875.652", "54873.253"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004090", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663099", "gt_frame_id": "54896.044", "anchor_frame_id": "54892.346", "shot": 6, "contexts_frame_id": ["54892.346", "54873.653", "54876.452", "54876.052", "54875.652", "54950.455"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004093", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663105", "gt_frame_id": "54785.057", "anchor_frame_id": "54760.450", "shot": 6, "contexts_frame_id": ["54760.450", "54781.958", "54776.760", "54746.956", "54753.953", "54749.755"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters, then move right about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004096", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663105", "gt_frame_id": "54757.351", "anchor_frame_id": "54781.958", "shot": 6, "contexts_frame_id": ["54781.958", "54782.957", "54746.356", "54744.757", "54787.755", "54799.351"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 1.2 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004099", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663105", "gt_frame_id": "54781.858", "anchor_frame_id": "54763.549", "shot": 6, "contexts_frame_id": ["54763.549", "54785.357", "54761.350", "54746.956", "54753.953", "54749.755"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move backward about 1.2 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004102", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663107", "gt_frame_id": "54812.145", "anchor_frame_id": "54808.647", "shot": 6, "contexts_frame_id": ["54808.647", "54809.047", "54807.048", "54824.257", "54835.652", "54862.058"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004105", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663149", "gt_frame_id": "63303.909", "anchor_frame_id": "63302.510", "shot": 6, "contexts_frame_id": ["63302.510", "63351.307", "63351.707", "63341.511", "63320.703", "63299.311"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004111", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663154", "gt_frame_id": "63452.459", "anchor_frame_id": "63431.450", "shot": 6, "contexts_frame_id": ["63431.450", "63431.051", "63463.654", "63430.851", "63429.851", "63428.852"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004114", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663164", "gt_frame_id": "65950.627", "anchor_frame_id": "65955.425", "shot": 6, "contexts_frame_id": ["65955.425", "65953.226", "65984.414", "65966.421", "65967.820", "66001.024"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move backward about 1.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004117", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663164", "gt_frame_id": "65991.428", "anchor_frame_id": "65988.929", "shot": 6, "contexts_frame_id": ["65988.929", "65955.425", "65953.626", "65952.426", "65950.827", "65968.220"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004120", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663164", "gt_frame_id": "65946.929", "anchor_frame_id": "65944.513", "shot": 6, "contexts_frame_id": ["65944.513", "66035.827", "66005.822", "65999.025", "65969.020", "66015.218"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004126", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663164", "gt_frame_id": "66017.617", "anchor_frame_id": "66008.021", "shot": 6, "contexts_frame_id": ["66008.021", "65969.020", "65946.029", "65940.414", "65943.813", "66026.414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004129", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663164", "gt_frame_id": "65988.429", "anchor_frame_id": "65991.528", "shot": 6, "contexts_frame_id": ["65991.528", "65956.025", "65953.026", "65951.027", "65968.220", "66007.022"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004132", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "45663164", "gt_frame_id": "66018.517", "anchor_frame_id": "66010.620", "shot": 6, "contexts_frame_id": ["66010.620", "65968.620", "66006.422", "65943.813", "65932.418", "65945.413"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004135", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663164", "gt_frame_id": "66007.121", "anchor_frame_id": "66017.617", "shot": 6, "contexts_frame_id": ["66017.617", "65946.829", "65968.820", "65968.420", "65967.420", "65997.026"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then move backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004138", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "45663175", "gt_frame_id": "65879.622", "anchor_frame_id": "65873.024", "shot": 6, "contexts_frame_id": ["65873.024", "65875.024", "65875.823", "65920.223", "65921.622", "65884.620"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004141", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115452", "gt_frame_id": "13069.779", "anchor_frame_id": "13064.481", "shot": 6, "contexts_frame_id": ["13064.481", "13113.578", "13115.977", "13086.572", "13087.772", "13088.571"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004144", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47115452", "gt_frame_id": "13066.980", "anchor_frame_id": "13093.369", "shot": 6, "contexts_frame_id": ["13093.369", "13062.182", "13087.572", "13063.182", "13113.378", "13072.978"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004147", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115452", "gt_frame_id": "13067.380", "anchor_frame_id": "13069.879", "shot": 6, "contexts_frame_id": ["13069.879", "13112.578", "13063.382", "13114.177", "13086.572", "13087.772"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004150", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115469", "gt_frame_id": "14014.392", "anchor_frame_id": "14017.891", "shot": 6, "contexts_frame_id": ["14017.891", "14040.681", "14045.679", "14017.491", "14021.689", "14023.688"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004153", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115525", "gt_frame_id": "14500.288", "anchor_frame_id": "14494.890", "shot": 6, "contexts_frame_id": ["14494.890", "14495.190", "14518.197", "14494.390", "14520.796", "14522.196"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004156", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47115525", "gt_frame_id": "14500.488", "anchor_frame_id": "14520.197", "shot": 6, "contexts_frame_id": ["14520.197", "14520.796", "14522.196", "14494.390", "14495.190", "14518.197"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.9 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004159", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47115525", "gt_frame_id": "14501.387", "anchor_frame_id": "14512.183", "shot": 6, "contexts_frame_id": ["14512.183", "14513.183", "14504.186", "14495.790", "14494.990", "14519.797"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004162", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115543", "gt_frame_id": "15633.211", "anchor_frame_id": "15630.912", "shot": 6, "contexts_frame_id": ["15630.912", "15627.613", "15628.612", "15665.814", "15666.414", "15661.616"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004165", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115543", "gt_frame_id": "15650.420", "anchor_frame_id": "15651.520", "shot": 6, "contexts_frame_id": ["15651.520", "15642.224", "15676.610", "15677.626", "15678.226", "15630.811"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004168", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47115543", "gt_frame_id": "15657.018", "anchor_frame_id": "15676.410", "shot": 6, "contexts_frame_id": ["15676.410", "15642.623", "15644.023", "15645.023", "15646.222", "15670.812"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around, then move backward about 1.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004171", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47115550", "gt_frame_id": "15857.522", "anchor_frame_id": "15884.428", "shot": 6, "contexts_frame_id": ["15884.428", "15883.112", "15872.516", "15871.716", "15868.917", "15867.918"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004174", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47204554", "gt_frame_id": "3268.296", "anchor_frame_id": "3266.197", "shot": 6, "contexts_frame_id": ["3266.197", "3214.201", "3262.398", "3273.194", "3261.998", "3246.788"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move forward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004177", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47204556", "gt_frame_id": "3316.193", "anchor_frame_id": "3317.692", "shot": 6, "contexts_frame_id": ["3317.692", "3321.990", "3322.790", "3296.001", "3338.200", "3338.800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004183", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47204559", "gt_frame_id": "3995.851", "anchor_frame_id": "3992.036", "shot": 6, "contexts_frame_id": ["3992.036", "3990.137", "3998.350", "4001.749", "4018.742", "4014.943"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004186", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47204566", "gt_frame_id": "4116.851", "anchor_frame_id": "4114.835", "shot": 6, "contexts_frame_id": ["4114.835", "4104.240", "4102.840", "4139.042", "4081.649", "4085.647"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004189", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47204566", "gt_frame_id": "4081.949", "anchor_frame_id": "4078.450", "shot": 6, "contexts_frame_id": ["4078.450", "4079.050", "4111.437", "4108.438", "4114.036", "4115.851"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004192", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47204573", "gt_frame_id": "5018.515", "anchor_frame_id": "5021.214", "shot": 6, "contexts_frame_id": ["5021.214", "4964.820", "4966.820", "4964.421", "4961.222", "5071.010"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move left about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004198", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47204575", "gt_frame_id": "4868.111", "anchor_frame_id": "4866.211", "shot": 6, "contexts_frame_id": ["4866.211", "4863.612", "4862.613", "4885.420", "4835.224", "4836.824"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004201", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47204578", "gt_frame_id": "4927.919", "anchor_frame_id": "4925.220", "shot": 6, "contexts_frame_id": ["4925.220", "4924.421", "4930.618", "4910.010", "4908.610", "4906.011"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004204", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47204609", "gt_frame_id": "2461.393", "anchor_frame_id": "2460.293", "shot": 6, "contexts_frame_id": ["2460.293", "2494.396", "2466.391", "2492.597", "2472.405", "2473.005"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004207", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47204609", "gt_frame_id": "2470.489", "anchor_frame_id": "2472.805", "shot": 6, "contexts_frame_id": ["2472.805", "2492.197", "2492.597", "2466.990", "2461.393", "2494.796"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004210", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331063", "gt_frame_id": "129469.052", "anchor_frame_id": "129494.941", "shot": 6, "contexts_frame_id": ["129494.941", "129494.542", "129490.343", "129489.743", "129464.354", "129439.147"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004213", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331064", "gt_frame_id": "129727.147", "anchor_frame_id": "129733.345", "shot": 6, "contexts_frame_id": ["129733.345", "129746.756", "129707.755", "129705.556", "129706.955", "129906.541"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004216", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331262", "gt_frame_id": "89944.995", "anchor_frame_id": "89946.194", "shot": 6, "contexts_frame_id": ["89946.194", "89938.897", "89949.093", "89940.896", "89939.497", "89923.903"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004219", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331262", "gt_frame_id": "89923.903", "anchor_frame_id": "89922.204", "shot": 6, "contexts_frame_id": ["89922.204", "89942.096", "89944.695", "89939.097", "89949.293", "89955.490"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004222", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331265", "gt_frame_id": "90020.497", "anchor_frame_id": "90018.598", "shot": 6, "contexts_frame_id": ["90018.598", "90061.897", "90016.099", "90013.700", "90062.297", "90029.694"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.7 meters, then move left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004225", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331266", "gt_frame_id": "89876.206", "anchor_frame_id": "89872.790", "shot": 6, "contexts_frame_id": ["89872.790", "89892.699", "89882.903", "89861.295", "89828.691", "89829.091"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004228", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331644", "gt_frame_id": "22193.113", "anchor_frame_id": "22218.603", "shot": 6, "contexts_frame_id": ["22218.603", "22210.906", "22206.308", "22204.508", "22201.310", "22164.908"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004231", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331644", "gt_frame_id": "22193.513", "anchor_frame_id": "22201.410", "shot": 6, "contexts_frame_id": ["22201.410", "22201.909", "22164.108", "22164.908", "22184.900", "22187.115"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004234", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331645", "gt_frame_id": "21893.901", "anchor_frame_id": "21883.506", "shot": 6, "contexts_frame_id": ["21883.506", "21869.811", "21881.007", "21867.612", "21899.816", "21900.615"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.3 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004237", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331646", "gt_frame_id": "22087.306", "anchor_frame_id": "22064.016", "shot": 6, "contexts_frame_id": ["22064.016", "22080.609", "22072.012", "22073.612", "21964.206", "22095.003"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004240", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47331646", "gt_frame_id": "22027.114", "anchor_frame_id": "22012.603", "shot": 6, "contexts_frame_id": ["22012.603", "21996.010", "22019.800", "22023.016", "22035.810", "22042.807"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004243", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331651", "gt_frame_id": "22918.324", "anchor_frame_id": "22906.529", "shot": 6, "contexts_frame_id": ["22906.529", "22914.826", "22908.628", "22909.428", "22929.236", "22934.234"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004246", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331651", "gt_frame_id": "22934.035", "anchor_frame_id": "22948.229", "shot": 6, "contexts_frame_id": ["22948.229", "22942.031", "22937.433", "22929.037", "22909.428", "22914.826"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004249", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331653", "gt_frame_id": "23110.429", "anchor_frame_id": "23095.335", "shot": 6, "contexts_frame_id": ["23095.335", "23099.534", "23117.126", "23119.525", "23126.323", "23130.721"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.3 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004255", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331654", "gt_frame_id": "22813.933", "anchor_frame_id": "22800.039", "shot": 6, "contexts_frame_id": ["22800.039", "22751.226", "22744.229", "22807.636", "22816.632", "22821.231"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004258", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331963", "gt_frame_id": "167656.541", "anchor_frame_id": "167665.437", "shot": 6, "contexts_frame_id": ["167665.437", "167661.139", "167663.138", "167785.338", "167778.941", "167698.140"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move right about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004261", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47331963", "gt_frame_id": "167731.244", "anchor_frame_id": "167734.042", "shot": 6, "contexts_frame_id": ["167734.042", "167734.742", "167812.944", "167811.744", "167741.139", "167726.546"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004264", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331964", "gt_frame_id": "167571.541", "anchor_frame_id": "167566.244", "shot": 6, "contexts_frame_id": ["167566.244", "167523.344", "167552.349", "167583.937", "167557.547", "167553.949"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004267", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331964", "gt_frame_id": "167587.552", "anchor_frame_id": "167592.150", "shot": 6, "contexts_frame_id": ["167592.150", "167553.949", "167558.147", "167552.749", "167570.342", "167574.940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters, then move backward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004273", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331964", "gt_frame_id": "167587.652", "anchor_frame_id": "167599.547", "shot": 6, "contexts_frame_id": ["167599.547", "167491.940", "167553.949", "167553.749", "167558.147", "167573.940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.3 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004279", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331970", "gt_frame_id": "167444.543", "anchor_frame_id": "167454.939", "shot": 6, "contexts_frame_id": ["167454.939", "167456.538", "167457.737", "167448.541", "167413.339", "167410.740"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004282", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331971", "gt_frame_id": "169241.268", "anchor_frame_id": "169238.869", "shot": 6, "contexts_frame_id": ["169238.869", "169235.771", "169232.972", "169141.976", "169142.575", "169146.774"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.4 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004288", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331971", "gt_frame_id": "169080.267", "anchor_frame_id": "169083.466", "shot": 6, "contexts_frame_id": ["169083.466", "169210.364", "169212.763", "169075.369", "169076.769", "169070.371"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004291", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331972", "gt_frame_id": "168680.279", "anchor_frame_id": "168687.177", "shot": 6, "contexts_frame_id": ["168687.177", "168716.781", "168717.581", "168723.378", "168724.578", "168735.174"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters, then move backward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004294", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331972", "gt_frame_id": "168706.169", "anchor_frame_id": "168709.268", "shot": 6, "contexts_frame_id": ["168709.268", "168701.571", "168699.971", "168699.571", "168726.177", "168731.575"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004315", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331974", "gt_frame_id": "168877.266", "anchor_frame_id": "168887.779", "shot": 6, "contexts_frame_id": ["168887.779", "168882.681", "168882.081", "168849.677", "168846.879", "168870.669"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.4 meters, then move backward about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004318", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331974", "gt_frame_id": "168835.367", "anchor_frame_id": "168863.772", "shot": 6, "contexts_frame_id": ["168863.772", "168857.674", "168838.865", "168840.681", "168846.879", "168894.476"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.0 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004321", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331974", "gt_frame_id": "168861.573", "anchor_frame_id": "168881.365", "shot": 6, "contexts_frame_id": ["168881.365", "168849.478", "168847.478", "168869.869", "168868.870", "168857.074"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.7 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004324", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331988", "gt_frame_id": "170733.651", "anchor_frame_id": "170708.161", "shot": 6, "contexts_frame_id": ["170708.161", "170702.763", "170711.160", "170738.349", "170736.750", "170716.958"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004327", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47331990", "gt_frame_id": "170528.251", "anchor_frame_id": "170536.164", "shot": 6, "contexts_frame_id": ["170536.164", "170561.354", "170532.349", "170559.755", "170523.353", "170516.555"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then move backward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004330", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47331990", "gt_frame_id": "170531.749", "anchor_frame_id": "170561.554", "shot": 6, "contexts_frame_id": ["170561.554", "170555.356", "170527.351", "170564.353", "170545.960", "170548.159"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004333", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332885", "gt_frame_id": "7500.459", "anchor_frame_id": "7516.269", "shot": 6, "contexts_frame_id": ["7516.269", "7517.669", "7494.861", "7509.272", "7508.672", "7507.256"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.6 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004336", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332890", "gt_frame_id": "7686.067", "anchor_frame_id": "7681.469", "shot": 6, "contexts_frame_id": ["7681.469", "7650.965", "7652.364", "7651.364", "7648.166", "7672.356"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.3 meters, then move forward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004345", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332893", "gt_frame_id": "9248.698", "anchor_frame_id": "9227.889", "shot": 6, "contexts_frame_id": ["9227.889", "9217.693", "9195.486", "9255.695", "9194.286", "9310.289"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004348", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332893", "gt_frame_id": "9230.288", "anchor_frame_id": "9212.796", "shot": 6, "contexts_frame_id": ["9212.796", "9201.683", "9246.898", "9217.693", "9195.686", "9319.885"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.0 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004351", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332893", "gt_frame_id": "9198.485", "anchor_frame_id": "9215.794", "shot": 6, "contexts_frame_id": ["9215.794", "9323.884", "9214.095", "9220.492", "9247.098", "9208.697"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004354", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332893", "gt_frame_id": "9220.792", "anchor_frame_id": "9249.697", "shot": 6, "contexts_frame_id": ["9249.697", "9193.087", "9311.489", "9195.486", "9198.485", "9246.898"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move right about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004357", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332895", "gt_frame_id": "9067.188", "anchor_frame_id": "9065.888", "shot": 6, "contexts_frame_id": ["9065.888", "9055.392", "9063.189", "9106.189", "9102.990", "9071.186"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004363", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47332899", "gt_frame_id": "9150.087", "anchor_frame_id": "9152.586", "shot": 6, "contexts_frame_id": ["9152.586", "9147.389", "9180.192", "9175.394", "9174.594", "9162.199"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004366", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47332899", "gt_frame_id": "9154.586", "anchor_frame_id": "9149.188", "shot": 6, "contexts_frame_id": ["9149.188", "9179.592", "9174.594", "9125.997", "9126.197", "9166.597"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004369", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332899", "gt_frame_id": "9131.095", "anchor_frame_id": "9127.597", "shot": 6, "contexts_frame_id": ["9127.597", "9128.596", "9142.591", "9139.992", "9139.592", "9174.594"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then turn right about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004375", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332904", "gt_frame_id": "11095.185", "anchor_frame_id": "11087.288", "shot": 6, "contexts_frame_id": ["11087.288", "11084.589", "11109.796", "11106.397", "11089.188", "11137.585"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004378", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332904", "gt_frame_id": "11147.097", "anchor_frame_id": "11138.301", "shot": 6, "contexts_frame_id": ["11138.301", "11130.188", "11088.188", "11090.387", "11141.999", "11139.800"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004381", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332905", "gt_frame_id": "10961.689", "anchor_frame_id": "10966.987", "shot": 6, "contexts_frame_id": ["10966.987", "10973.301", "10964.688", "10956.691", "10986.896", "10981.298"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move backward about 0.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004384", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332905", "gt_frame_id": "10968.386", "anchor_frame_id": "10984.297", "shot": 6, "contexts_frame_id": ["10984.297", "10964.488", "10972.301", "10978.899", "10975.900", "10970.885"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.8 meters, then move backward about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004387", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47332905", "gt_frame_id": "10966.987", "anchor_frame_id": "10961.689", "shot": 6, "contexts_frame_id": ["10961.689", "10964.288", "10973.501", "10975.900", "10978.899", "10970.885"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.9 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004390", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47332915", "gt_frame_id": "13508.273", "anchor_frame_id": "13511.072", "shot": 6, "contexts_frame_id": ["13511.072", "13474.370", "13473.171", "13546.574", "13525.966", "13519.568"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004393", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47332918", "gt_frame_id": "13336.576", "anchor_frame_id": "13342.274", "shot": 6, "contexts_frame_id": ["13342.274", "13325.464", "13324.864", "13364.465", "13364.865", "13366.464"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004396", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333441", "gt_frame_id": "53878.229", "anchor_frame_id": "53860.020", "shot": 6, "contexts_frame_id": ["53860.020", "53859.920", "53859.520", "53868.916", "53871.931", "53871.132"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters, then move forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004399", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333452", "gt_frame_id": "54962.819", "anchor_frame_id": "54958.220", "shot": 6, "contexts_frame_id": ["54958.220", "54969.416", "54970.615", "54957.921", "54957.521", "54956.121"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004402", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333452", "gt_frame_id": "54964.718", "anchor_frame_id": "54970.215", "shot": 6, "contexts_frame_id": ["54970.215", "54971.415", "54957.921", "54957.521", "54956.121", "54955.322"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters, then move left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004408", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333457", "gt_frame_id": "54913.422", "anchor_frame_id": "54909.424", "shot": 6, "contexts_frame_id": ["54909.424", "54910.124", "54906.308", "54905.109", "54902.110", "54901.311"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004411", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333899", "gt_frame_id": "81300.868", "anchor_frame_id": "81283.974", "shot": 6, "contexts_frame_id": ["81283.974", "81286.274", "81284.474", "81281.076", "81279.276", "81274.678"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004414", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333899", "gt_frame_id": "81286.673", "anchor_frame_id": "81300.868", "shot": 6, "contexts_frame_id": ["81300.868", "81302.467", "81289.672", "81303.467", "81303.866", "81250.072"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters, then move forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004417", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333916", "gt_frame_id": "82378.753", "anchor_frame_id": "82379.952", "shot": 6, "contexts_frame_id": ["82379.952", "82376.154", "82420.353", "82375.354", "82421.152", "82427.350"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004420", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333918", "gt_frame_id": "82470.766", "anchor_frame_id": "82474.764", "shot": 6, "contexts_frame_id": ["82474.764", "82478.662", "82463.852", "82462.053", "82508.850", "82512.665"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004429", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333918", "gt_frame_id": "82473.465", "anchor_frame_id": "82466.851", "shot": 6, "contexts_frame_id": ["82466.851", "82476.463", "82492.057", "82478.262", "82489.658", "82524.660"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004432", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47333920", "gt_frame_id": "82564.661", "anchor_frame_id": "82577.056", "shot": 6, "contexts_frame_id": ["82577.056", "82581.854", "82588.251", "82608.260", "82579.055", "82647.260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004435", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333920", "gt_frame_id": "82576.356", "anchor_frame_id": "82602.662", "shot": 6, "contexts_frame_id": ["82602.662", "82594.465", "82605.661", "82610.259", "82565.460", "82564.061"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004438", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333925", "gt_frame_id": "83578.111", "anchor_frame_id": "83583.409", "shot": 6, "contexts_frame_id": ["83583.409", "83584.609", "83547.807", "83556.603", "83558.003", "83558.403"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 75 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004441", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333934", "gt_frame_id": "85585.877", "anchor_frame_id": "85556.372", "shot": 6, "contexts_frame_id": ["85556.372", "85558.371", "85555.772", "85637.772", "85595.373", "85593.573"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004444", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47333934", "gt_frame_id": "85588.675", "anchor_frame_id": "85575.664", "shot": 6, "contexts_frame_id": ["85575.664", "85583.578", "85555.772", "85556.372", "85558.771", "85558.371"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004447", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333934", "gt_frame_id": "85663.862", "anchor_frame_id": "85657.564", "shot": 6, "contexts_frame_id": ["85657.564", "85595.373", "85568.167", "85654.765", "85640.771", "85600.970"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004450", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333937", "gt_frame_id": "85539.762", "anchor_frame_id": "85541.978", "shot": 6, "contexts_frame_id": ["85541.978", "85508.974", "85510.574", "85507.775", "85496.563", "85534.364"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004459", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333937", "gt_frame_id": "85497.063", "anchor_frame_id": "85526.367", "shot": 6, "contexts_frame_id": ["85526.367", "85531.365", "85504.776", "85508.175", "85505.176", "85506.176"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.0 meters, then turn right about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004462", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333940", "gt_frame_id": "85443.668", "anchor_frame_id": "85436.470", "shot": 6, "contexts_frame_id": ["85436.470", "85417.978", "85419.178", "85449.765", "85450.965", "85452.364"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters, then move left about 1.1 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004465", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333940", "gt_frame_id": "85434.471", "anchor_frame_id": "85440.669", "shot": 6, "contexts_frame_id": ["85440.669", "85458.778", "85459.978", "85418.977", "85420.777", "85452.764"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.1 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004468", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47333940", "gt_frame_id": "85443.867", "anchor_frame_id": "85462.377", "shot": 6, "contexts_frame_id": ["85462.377", "85463.976", "85458.778", "85418.778", "85419.178", "85420.977"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 90 degrees, then move left about 0.8 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004471", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47333940", "gt_frame_id": "85437.970", "anchor_frame_id": "85444.567", "shot": 6, "contexts_frame_id": ["85444.567", "85418.977", "85417.578", "85458.978", "85462.777", "85465.975"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.7 meters, then move right about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004474", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334093", "gt_frame_id": "91936.027", "anchor_frame_id": "91931.929", "shot": 6, "contexts_frame_id": ["91931.929", "91921.633", "91922.432", "91923.232", "91949.838", "91847.229"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.6 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004480", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334093", "gt_frame_id": "91932.029", "anchor_frame_id": "91935.827", "shot": 6, "contexts_frame_id": ["91935.827", "91922.632", "91922.033", "91918.034", "91910.037", "91909.038"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004483", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334102", "gt_frame_id": "92853.448", "anchor_frame_id": "92857.347", "shot": 6, "contexts_frame_id": ["92857.347", "92866.360", "92862.145", "92830.957", "92845.551", "92844.352"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn right about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004489", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334105", "gt_frame_id": "92913.057", "anchor_frame_id": "92909.259", "shot": 6, "contexts_frame_id": ["92909.259", "92903.745", "92919.755", "92920.754", "92922.354", "92951.159"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004495", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334105", "gt_frame_id": "93033.159", "anchor_frame_id": "93015.949", "shot": 6, "contexts_frame_id": ["93015.949", "93018.148", "93029.743", "93039.156", "93038.956", "93038.756"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 2.0 meters, then move forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004501", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334105", "gt_frame_id": "92909.159", "anchor_frame_id": "92901.345", "shot": 6, "contexts_frame_id": ["92901.345", "92898.746", "92919.755", "92922.154", "92914.357", "93043.754"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.3 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004504", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334110", "gt_frame_id": "93580.902", "anchor_frame_id": "93560.394", "shot": 6, "contexts_frame_id": ["93560.394", "93562.893", "93577.887", "93534.488", "93539.302", "93661.286"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004507", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334115", "gt_frame_id": "94138.547", "anchor_frame_id": "94153.057", "shot": 6, "contexts_frame_id": ["94153.057", "94127.751", "94127.351", "94128.351", "94156.756", "94160.354"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters, then turn left about 45 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004510", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334115", "gt_frame_id": "94173.649", "anchor_frame_id": "94164.553", "shot": 6, "contexts_frame_id": ["94164.553", "94169.751", "94169.151", "94178.547", "94186.544", "94190.142"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.5 meters, then move right about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004513", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334115", "gt_frame_id": "94169.351", "anchor_frame_id": "94165.952", "shot": 6, "contexts_frame_id": ["94165.952", "94164.153", "94172.749", "94173.549", "94178.747", "94186.544"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004516", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334115", "gt_frame_id": "94137.147", "anchor_frame_id": "94128.351", "shot": 6, "contexts_frame_id": ["94128.351", "94143.345", "94127.351", "94127.151", "94156.756", "94160.354"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move right about 2.1 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004519", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334117", "gt_frame_id": "94251.351", "anchor_frame_id": "94246.953", "shot": 6, "contexts_frame_id": ["94246.953", "94247.053", "94243.254", "94243.654", "94241.055", "94261.647"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004522", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334120", "gt_frame_id": "94357.442", "anchor_frame_id": "94327.853", "shot": 6, "contexts_frame_id": ["94327.853", "94315.842", "94315.442", "94338.849", "94343.047", "94344.447"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees, then move forward about 1.7 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004525", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334234", "gt_frame_id": "109741.583", "anchor_frame_id": "109749.180", "shot": 6, "contexts_frame_id": ["109749.180", "109627.179", "109746.981", "109681.374", "109704.582", "109696.368"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004528", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334234", "gt_frame_id": "109688.072", "anchor_frame_id": "109687.172", "shot": 6, "contexts_frame_id": ["109687.172", "109679.575", "109703.182", "109891.972", "109647.171", "109736.769"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004534", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334237", "gt_frame_id": "109542.580", "anchor_frame_id": "109540.881", "shot": 6, "contexts_frame_id": ["109540.881", "109517.874", "109587.879", "109593.676", "109524.271", "109507.678"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004543", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334238", "gt_frame_id": "109396.773", "anchor_frame_id": "109411.383", "shot": 6, "contexts_frame_id": ["109411.383", "109421.080", "109371.083", "109392.474", "109390.475", "109388.076"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.1 meters, then move left about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004546", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334238", "gt_frame_id": "109299.779", "anchor_frame_id": "109289.883", "shot": 6, "contexts_frame_id": ["109289.883", "109291.482", "109312.473", "109317.471", "109317.271", "109422.879"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004549", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334238", "gt_frame_id": "109401.571", "anchor_frame_id": "109389.875", "shot": 6, "contexts_frame_id": ["109389.875", "109393.474", "109396.273", "109421.080", "109370.283", "109405.069"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move forward about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004552", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334240", "gt_frame_id": "111169.014", "anchor_frame_id": "111180.326", "shot": 6, "contexts_frame_id": ["111180.326", "111184.424", "111185.024", "111177.827", "111178.627", "111174.228"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004555", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334252", "gt_frame_id": "112185.784", "anchor_frame_id": "112181.685", "shot": 6, "contexts_frame_id": ["112181.685", "112089.189", "112194.597", "112087.590", "112087.790", "112230.999"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004558", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334360", "gt_frame_id": "62049.727", "anchor_frame_id": "62043.329", "shot": 6, "contexts_frame_id": ["62043.329", "62054.325", "62060.123", "61996.315", "62026.720", "62010.726"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.7 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004561", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334360", "gt_frame_id": "62043.329", "anchor_frame_id": "62035.016", "shot": 6, "contexts_frame_id": ["62035.016", "61955.515", "61995.316", "62049.927", "61997.715", "62010.726"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004564", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334361", "gt_frame_id": "61794.930", "anchor_frame_id": "61787.516", "shot": 6, "contexts_frame_id": ["61787.516", "61803.627", "61798.429", "61799.828", "61728.824", "61727.424"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.2 meters, then move forward about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004567", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334362", "gt_frame_id": "61871.516", "anchor_frame_id": "61881.528", "shot": 6, "contexts_frame_id": ["61881.528", "61865.918", "61878.530", "61833.715", "61836.130", "61889.325"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004570", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334362", "gt_frame_id": "61871.516", "anchor_frame_id": "61852.724", "shot": 6, "contexts_frame_id": ["61852.724", "61867.317", "61855.322", "61865.118", "61836.130", "61886.926"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.7 meters, then move backward about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004573", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334379", "gt_frame_id": "64548.100", "anchor_frame_id": "64572.091", "shot": 6, "contexts_frame_id": ["64572.091", "64570.691", "64584.685", "64714.499", "64539.887", "64541.287"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then turn right about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004579", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47334380", "gt_frame_id": "64347.499", "anchor_frame_id": "64331.389", "shot": 6, "contexts_frame_id": ["64331.389", "64338.086", "64330.489", "64326.491", "64325.491", "64325.291"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004582", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334380", "gt_frame_id": "64331.289", "anchor_frame_id": "64332.988", "shot": 6, "contexts_frame_id": ["64332.988", "64322.492", "64326.690", "64354.296", "64352.097", "64325.091"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees, then move left about 0.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004585", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334381", "gt_frame_id": "64452.989", "anchor_frame_id": "64424.601", "shot": 6, "contexts_frame_id": ["64424.601", "64400.794", "64387.399", "64411.789", "64407.191", "64445.592"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters, then turn around.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004588", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47334381", "gt_frame_id": "64445.092", "anchor_frame_id": "64430.798", "shot": 6, "contexts_frame_id": ["64430.798", "64413.788", "64414.188", "64408.191", "64426.000", "64421.985"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees, then move forward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004591", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895341", "gt_frame_id": "96410.022", "anchor_frame_id": "96405.024", "shot": 6, "contexts_frame_id": ["96405.024", "96450.322", "96450.522", "96442.725", "96398.326", "96438.727"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters, then move backward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004594", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47895341", "gt_frame_id": "96412.221", "anchor_frame_id": "96394.228", "shot": 6, "contexts_frame_id": ["96394.228", "96439.527", "96450.922", "96450.722", "96450.522", "96450.322"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move right about 2.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004597", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47895348", "gt_frame_id": "96516.112", "anchor_frame_id": "96519.527", "shot": 6, "contexts_frame_id": ["96519.527", "96523.326", "96510.115", "96509.915", "96506.516", "96503.917"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move left about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004600", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895348", "gt_frame_id": "96518.528", "anchor_frame_id": "96506.916", "shot": 6, "contexts_frame_id": ["96506.916", "96510.515", "96509.915", "96533.522", "96533.722", "96524.725"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.4 meters, then move forward about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004606", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47895350", "gt_frame_id": "96618.321", "anchor_frame_id": "96635.114", "shot": 6, "contexts_frame_id": ["96635.114", "96598.812", "96596.413", "96730.825", "96572.823", "96602.827"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004609", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895350", "gt_frame_id": "96677.514", "anchor_frame_id": "96680.812", "shot": 6, "contexts_frame_id": ["96680.812", "96686.627", "96692.424", "96727.227", "96726.227", "96725.428"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.8 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004612", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47895350", "gt_frame_id": "96635.114", "anchor_frame_id": "96618.421", "shot": 6, "contexts_frame_id": ["96618.421", "96599.012", "96642.228", "96730.825", "96754.216", "96747.618"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 60 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004615", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895350", "gt_frame_id": "96667.818", "anchor_frame_id": "96684.927", "shot": 6, "contexts_frame_id": ["96684.927", "96678.213", "96726.227", "96727.227", "96691.825", "96654.823"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 1.4 meters, then move forward about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004624", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895353", "gt_frame_id": "98691.829", "anchor_frame_id": "98697.126", "shot": 6, "contexts_frame_id": ["98697.126", "98694.827", "98664.840", "98723.232", "98705.040", "98713.836"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.7 meters, then move backward about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004627", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "47895355", "gt_frame_id": "98831.338", "anchor_frame_id": "98848.731", "shot": 6, "contexts_frame_id": ["98848.731", "98945.625", "98808.831", "98811.430", "98814.829", "98876.637"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004630", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895355", "gt_frame_id": "98937.229", "anchor_frame_id": "98918.836", "shot": 6, "contexts_frame_id": ["98918.836", "98931.631", "98933.030", "98954.638", "98965.634", "98959.636"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.5 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004633", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47895364", "gt_frame_id": "100314.637", "anchor_frame_id": "100297.444", "shot": 6, "contexts_frame_id": ["100297.444", "100296.344", "100293.745", "100279.934", "100310.339", "100274.537"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004636", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895367", "gt_frame_id": "100042.731", "anchor_frame_id": "100050.344", "shot": 6, "contexts_frame_id": ["100050.344", "100046.845", "100091.444", "100090.644", "100056.242", "100036.633"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004642", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47895371", "gt_frame_id": "102085.830", "anchor_frame_id": "102072.435", "shot": 6, "contexts_frame_id": ["102072.435", "102074.235", "102074.635", "102075.034", "102111.036", "102081.632"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 45 degrees, then move left about 1.4 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004645", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "47895374", "gt_frame_id": "101817.439", "anchor_frame_id": "101821.437", "shot": 6, "contexts_frame_id": ["101821.437", "101827.835", "101834.632", "101833.632", "101832.633", "101812.641"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.6 meters, then move forward about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004648", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "47895374", "gt_frame_id": "101876.632", "anchor_frame_id": "101881.030", "shot": 6, "contexts_frame_id": ["101881.030", "101881.629", "101879.630", "101872.633", "101868.635", "101870.034"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.0 meters, then turn left about 75 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004651", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018345", "gt_frame_id": "388678.935", "anchor_frame_id": "388704.841", "shot": 6, "contexts_frame_id": ["388704.841", "388671.438", "388670.639", "388674.037", "388674.237", "388701.043"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004654", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48018353", "gt_frame_id": "389135.247", "anchor_frame_id": "389136.346", "shot": 6, "contexts_frame_id": ["389136.346", "389125.651", "389127.250", "389129.449", "389130.249", "389150.857"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees, then move backward about 0.6 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004660", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018356", "gt_frame_id": "388776.945", "anchor_frame_id": "388786.941", "shot": 6, "contexts_frame_id": ["388786.941", "388825.941", "388830.340", "388793.738", "388791.139", "388790.539"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.2 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004663", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48018360", "gt_frame_id": "389177.346", "anchor_frame_id": "389173.947", "shot": 6, "contexts_frame_id": ["389173.947", "389185.143", "389184.743", "389183.943", "389183.144", "389179.945"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters, then turn left about 90 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004666", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48018361", "gt_frame_id": "389015.546", "anchor_frame_id": "389011.448", "shot": 6, "contexts_frame_id": ["389011.448", "389020.544", "389043.151", "389019.544", "389042.152", "389041.552"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.1 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004669", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018375", "gt_frame_id": "390079.836", "anchor_frame_id": "390084.534", "shot": 6, "contexts_frame_id": ["390084.534", "390150.840", "390134.430", "390069.840", "390130.032", "390087.233"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.6 meters, then move right about 0.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004672", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018381", "gt_frame_id": "389965.033", "anchor_frame_id": "389993.438", "shot": 6, "contexts_frame_id": ["389993.438", "389947.241", "389969.232", "389972.030", "389973.430", "389969.831"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.3 meters, then move right about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004675", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018386", "gt_frame_id": "390438.649", "anchor_frame_id": "390434.534", "shot": 6, "contexts_frame_id": ["390434.534", "390434.334", "390426.937", "390432.735", "390460.540", "390458.940"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.6 meters, then move left about 0.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004678", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018566", "gt_frame_id": "3864.236", "anchor_frame_id": "3865.835", "shot": 6, "contexts_frame_id": ["3865.835", "3859.838", "3879.430", "3878.630", "3878.430", "3888.643"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004681", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018572", "gt_frame_id": "3798.946", "anchor_frame_id": "3805.143", "shot": 6, "contexts_frame_id": ["3805.143", "3803.444", "3807.042", "3834.231", "3825.835", "3831.033"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.7 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004687", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48018741", "gt_frame_id": "3390.626", "anchor_frame_id": "3420.614", "shot": 6, "contexts_frame_id": ["3420.614", "3417.415", "3417.615", "3397.023", "3402.221", "3396.424"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004690", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "48458415", "gt_frame_id": "115217.146", "anchor_frame_id": "115212.248", "shot": 6, "contexts_frame_id": ["115212.248", "115205.351", "115204.552", "115191.357", "115182.761", "115183.160"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004693", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458417", "gt_frame_id": "115313.256", "anchor_frame_id": "115311.557", "shot": 6, "contexts_frame_id": ["115311.557", "115320.353", "115321.753", "115306.759", "115327.950", "115336.747"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.3 meters, then move right about 0.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004696", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458417", "gt_frame_id": "115313.956", "anchor_frame_id": "115322.353", "shot": 6, "contexts_frame_id": ["115322.353", "115311.357", "115319.754", "115306.759", "115327.950", "115336.747"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 0.9 meters, then turn left about 60 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004705", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458424", "gt_frame_id": "115120.353", "anchor_frame_id": "115145.659", "shot": 6, "contexts_frame_id": ["115145.659", "115107.658", "115146.659", "115127.050", "115151.657", "115117.254"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004708", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458424", "gt_frame_id": "115157.954", "anchor_frame_id": "115156.355", "shot": 6, "contexts_frame_id": ["115156.355", "115130.849", "115131.249", "115149.658", "115120.253", "115109.857"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.5 meters, then turn right about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004711", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458424", "gt_frame_id": "115145.359", "anchor_frame_id": "115169.149", "shot": 6, "contexts_frame_id": ["115169.149", "115166.651", "115164.052", "115163.252", "115110.057", "115142.860"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn around, then move backward about 0.5 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004717", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "48458424", "gt_frame_id": "115117.354", "anchor_frame_id": "115118.554", "shot": 6, "contexts_frame_id": ["115118.554", "115124.251", "115154.655", "115120.053", "115131.249", "115159.653"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004720", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458428", "gt_frame_id": "116476.247", "anchor_frame_id": "116472.849", "shot": 6, "contexts_frame_id": ["116472.849", "116471.949", "116472.749", "116491.941", "116498.538", "116499.738"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.4 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004723", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458465", "gt_frame_id": "110491.151", "anchor_frame_id": "110484.354", "shot": 6, "contexts_frame_id": ["110484.354", "110488.152", "110494.150", "110485.753", "110509.744", "110508.544"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 3.0 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004726", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458472", "gt_frame_id": "110532.651", "anchor_frame_id": "110527.154", "shot": 6, "contexts_frame_id": ["110527.154", "110529.952", "110529.753", "110589.046", "110580.149", "110578.950"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.6 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004729", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458481", "gt_frame_id": "111510.506", "anchor_frame_id": "111505.408", "shot": 6, "contexts_frame_id": ["111505.408", "111507.507", "111506.508", "111525.300", "111524.300", "111513.705"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.2 meters, then turn left about 30 degrees.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004732", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458647", "gt_frame_id": "7018.705", "anchor_frame_id": "7047.010", "shot": 6, "contexts_frame_id": ["7047.010", "7022.704", "7023.104", "7027.102", "7012.508", "7044.711"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove backward about 1.3 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004738", "dataset": "arkitscenes", "task_type": "rotation", "scene_id": "48458647", "gt_frame_id": "7048.010", "anchor_frame_id": "7048.810", "shot": 6, "contexts_frame_id": ["7048.810", "7032.716", "7029.118", "7018.905", "7028.118", "7017.906"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 30 degrees.\n\nKeep the camera center, tilt, roll, and field of view fixed; change only the horizontal heading.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004741", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458650", "gt_frame_id": "7110.617", "anchor_frame_id": "7094.707", "shot": 6, "contexts_frame_id": ["7094.707", "7104.603", "7095.407", "7066.002", "7068.202", "7073.216"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.5 meters, then turn around.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004744", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458650", "gt_frame_id": "7107.502", "anchor_frame_id": "7078.014", "shot": 6, "contexts_frame_id": ["7078.014", "7066.402", "7076.615", "7100.005", "7099.006", "7098.206"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 0.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004747", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458663", "gt_frame_id": "8327.854", "anchor_frame_id": "8317.159", "shot": 6, "contexts_frame_id": ["8317.159", "8315.260", "8317.658", "8319.058", "8341.849", "8344.447"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 2.8 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004750", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458735", "gt_frame_id": "139225.589", "anchor_frame_id": "139241.699", "shot": 6, "contexts_frame_id": ["139241.699", "139183.089", "139182.489", "139208.495", "139172.293", "139251.895"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn left about 45 degrees, then move right about 1.9 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004756", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458735", "gt_frame_id": "139225.888", "anchor_frame_id": "139237.684", "shot": 6, "contexts_frame_id": ["139237.684", "139230.687", "139183.089", "139182.489", "139208.895", "139241.899"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove right about 1.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004759", "dataset": "arkitscenes", "task_type": "translation", "scene_id": "48458747", "gt_frame_id": "139055.489", "anchor_frame_id": "139041.295", "shot": 6, "contexts_frame_id": ["139041.295", "139067.084", "139069.284", "139041.495", "139074.698", "139073.682"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove forward about 1.9 meters, then move left about 0.9 meters.\n\nMovement directions are relative to the current camera heading. Translate without changing heading, height, roll, or field of view.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004765", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458747", "gt_frame_id": "139057.988", "anchor_frame_id": "139075.498", "shot": 6, "contexts_frame_id": ["139075.498", "139031.899", "139069.883", "139041.695", "139051.091", "139086.294"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nMove left about 0.5 meters, then turn around.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004768", "dataset": "arkitscenes", "task_type": "composition", "scene_id": "48458747", "gt_frame_id": "139058.088", "anchor_frame_id": "139061.487", "shot": 6, "contexts_frame_id": ["139061.487", "139072.283", "139063.086", "139086.294", "139051.091", "139041.695"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nCamera motion from Image 1:\nTurn right about 75 degrees, then move backward about 0.3 meters.\n\nFollow the written order. Each translation is relative to the camera heading at that step. Change only the stated translation and heading; keep height, tilt, roll, and field of view fixed.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004771", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "41159557", "gt_frame_id": "33263.151", "anchor_frame_id": "33251.256", "shot": 6, "contexts_frame_id": ["33251.256", "33252.055", "33277.862", "33285.859", "33260.652", "33310.849"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the shelf, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the shelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004774", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "42446056", "gt_frame_id": "349303.872", "anchor_frame_id": "349301.573", "shot": 6, "contexts_frame_id": ["349301.573", "349289.978", "349255.575", "349295.775", "349349.187", "349380.574"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004777", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "42446114", "gt_frame_id": "78254.693", "anchor_frame_id": "78257.991", "shot": 6, "contexts_frame_id": ["78257.991", "78247.396", "78365.982", "78367.581", "78276.584", "78274.985"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the shelf, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the shelf, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004780", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "42897559", "gt_frame_id": "277446.795", "anchor_frame_id": "277455.208", "shot": 6, "contexts_frame_id": ["277455.208", "277453.992", "277451.193", "277457.607", "277438.998", "277471.402"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004783", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "42897688", "gt_frame_id": "479241.791", "anchor_frame_id": "479234.077", "shot": 6, "contexts_frame_id": ["479234.077", "479246.289", "479251.887", "479254.485", "479261.682", "479266.281"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the stool, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the stool, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004786", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "42898869", "gt_frame_id": "197854.937", "anchor_frame_id": "197821.634", "shot": 6, "contexts_frame_id": ["197821.634", "197776.635", "197782.233", "197827.032", "197789.430", "197790.429"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004789", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "44358515", "gt_frame_id": "52574.718", "anchor_frame_id": "52592.410", "shot": 6, "contexts_frame_id": ["52592.410", "52593.310", "52610.719", "52615.317", "52583.514", "52572.119"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004792", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45261133", "gt_frame_id": "4096.361", "anchor_frame_id": "4097.661", "shot": 6, "contexts_frame_id": ["4097.661", "4038.869", "4104.258", "4034.670", "4089.664", "4013.063"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004795", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45261190", "gt_frame_id": "3715.986", "anchor_frame_id": "3746.791", "shot": 6, "contexts_frame_id": ["3746.791", "3748.290", "3707.689", "3741.892", "3740.693", "3723.283"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004798", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45261619", "gt_frame_id": "55132.593", "anchor_frame_id": "55086.677", "shot": 6, "contexts_frame_id": ["55086.677", "55088.893", "55095.291", "55079.080", "55078.281", "55076.881"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 45 degrees to the right around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004801", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45261619", "gt_frame_id": "55086.478", "anchor_frame_id": "55132.593", "shot": 6, "contexts_frame_id": ["55132.593", "55094.491", "55080.480", "55090.692", "55097.090", "55098.289"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004804", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45662926", "gt_frame_id": "21704.229", "anchor_frame_id": "21690.318", "shot": 6, "contexts_frame_id": ["21690.318", "21689.718", "21691.318", "21701.130", "21687.519", "21682.721"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004807", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45662983", "gt_frame_id": "27842.887", "anchor_frame_id": "27816.197", "shot": 6, "contexts_frame_id": ["27816.197", "27815.298", "27812.082", "27810.683", "27822.695", "27833.890"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004810", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45663149", "gt_frame_id": "63350.907", "anchor_frame_id": "63305.009", "shot": 6, "contexts_frame_id": ["63305.009", "63302.710", "63301.510", "63300.311", "63334.714", "63316.704"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the fireplace, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the fireplace, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004813", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45663149", "gt_frame_id": "63350.907", "anchor_frame_id": "63304.209", "shot": 6, "contexts_frame_id": ["63304.209", "63301.510", "63300.511", "63341.511", "63307.908", "63309.907"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the fireplace, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the fireplace, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004816", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45663150", "gt_frame_id": "63582.857", "anchor_frame_id": "63537.558", "shot": 6, "contexts_frame_id": ["63537.558", "63577.759", "63568.363", "63596.152", "63559.750", "63516.550"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004819", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45663154", "gt_frame_id": "63414.657", "anchor_frame_id": "63440.764", "shot": 6, "contexts_frame_id": ["63440.764", "63412.058", "63407.660", "63436.249", "63458.656", "63433.050"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004822", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "45663175", "gt_frame_id": "65889.818", "anchor_frame_id": "65917.624", "shot": 6, "contexts_frame_id": ["65917.624", "65901.413", "65906.628", "65898.015", "65910.027", "65912.625"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the refrigerator, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the refrigerator, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004825", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47204607", "gt_frame_id": "2669.894", "anchor_frame_id": "2668.594", "shot": 6, "contexts_frame_id": ["2668.594", "2528.200", "2655.199", "2644.803", "2679.007", "2636.590"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004828", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47204607", "gt_frame_id": "2668.594", "anchor_frame_id": "2669.594", "shot": 6, "contexts_frame_id": ["2669.594", "2664.596", "2528.399", "2653.800", "2680.006", "2680.806"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004831", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47204607", "gt_frame_id": "2666.995", "anchor_frame_id": "2669.594", "shot": 6, "contexts_frame_id": ["2669.594", "2663.796", "2527.400", "2659.798", "2679.007", "2680.206"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004834", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47331970", "gt_frame_id": "167446.242", "anchor_frame_id": "167448.841", "shot": 6, "contexts_frame_id": ["167448.841", "167443.543", "167413.139", "167453.739", "167457.338", "167457.737"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the chair, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the chair, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004837", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47331971", "gt_frame_id": "169199.569", "anchor_frame_id": "169183.275", "shot": 6, "contexts_frame_id": ["169183.275", "169177.178", "169193.771", "169162.367", "169165.166", "169171.164"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004840", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47331971", "gt_frame_id": "169165.066", "anchor_frame_id": "169183.275", "shot": 6, "contexts_frame_id": ["169183.275", "169181.376", "169169.564", "169193.571", "169151.772", "169150.572"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, to the opposite side of the object from the anchor view. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004843", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47331971", "gt_frame_id": "168978.875", "anchor_frame_id": "169203.767", "shot": 6, "contexts_frame_id": ["169203.767", "169179.777", "169182.576", "169175.778", "169258.578", "169259.178"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 90 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004846", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47332916", "gt_frame_id": "13440.567", "anchor_frame_id": "13414.878", "shot": 6, "contexts_frame_id": ["13414.878", "13399.168", "13397.568", "13436.569", "13402.166", "13407.764"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004849", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47332916", "gt_frame_id": "13414.878", "anchor_frame_id": "13440.567", "shot": 6, "contexts_frame_id": ["13440.567", "13400.167", "13397.168", "13433.570", "13409.963", "13431.171"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 30 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004852", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47332918", "gt_frame_id": "13323.565", "anchor_frame_id": "13366.364", "shot": 6, "contexts_frame_id": ["13366.364", "13365.065", "13340.875", "13338.076", "13381.275", "13373.678"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 90 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004855", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47333456", "gt_frame_id": "55030.207", "anchor_frame_id": "55032.223", "shot": 6, "contexts_frame_id": ["55032.223", "55034.722", "55035.522", "55036.321", "55037.121", "55021.311"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the cabinet, which is near the center of the first image, about 45 degrees to the left around the object. Keep a similar distance from the cabinet, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004858", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47333934", "gt_frame_id": "85639.671", "anchor_frame_id": "85574.465", "shot": 6, "contexts_frame_id": ["85574.465", "85600.970", "85634.773", "85555.772", "85628.176", "85558.571"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004861", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47333934", "gt_frame_id": "85641.071", "anchor_frame_id": "85601.270", "shot": 6, "contexts_frame_id": ["85601.270", "85636.573", "85580.779", "85553.773", "85653.966", "85571.966"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004864", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47895350", "gt_frame_id": "96751.217", "anchor_frame_id": "96761.113", "shot": 6, "contexts_frame_id": ["96761.113", "96753.816", "96763.412", "96574.422", "96592.415", "96615.222"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the right around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004867", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "47895353", "gt_frame_id": "98725.132", "anchor_frame_id": "98712.637", "shot": 6, "contexts_frame_id": ["98712.637", "98706.239", "98701.441", "98729.230", "98715.236", "98720.833"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the table, which is near the center of the first image, about 75 degrees to the left around the object. Keep a similar distance from the table, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004870", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "48018732", "gt_frame_id": "1923.471", "anchor_frame_id": "1924.471", "shot": 6, "contexts_frame_id": ["1924.471", "1967.970", "1979.765", "1928.569", "1990.577", "1937.765"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the sofa, which is near the center of the first image, about 30 degrees to the right around the object. Keep a similar distance from the sofa, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004873", "dataset": "arkitscenes", "task_type": "anchor_guided_relative_repositioning", "scene_id": "48458424", "gt_frame_id": "115159.353", "anchor_frame_id": "115114.655", "shot": 6, "contexts_frame_id": ["115114.655", "115129.849", "115130.649", "115131.249", "115111.856", "115111.257"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nObject-centered motion from Image 1:\nMove around the refrigerator, which is near the center of the first image, about 60 degrees to the left around the object. Keep a similar distance from the refrigerator, face it, and keep it near the center of the image.\n\nTreat left or right as the camera's orbit direction around the object, not an image shift. Obey the stated distance, facing, and centering constraints.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004876", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "41125756", "gt_frame_id": "5289.306", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["5319.910", "5315.312", "5295.303", "5294.703", "5293.104", "5302.700"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe toilet to the left of the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004879", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "41159534", "gt_frame_id": "39069.912", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["39028.212", "39072.810", "39074.210", "39026.013", "39023.614", "39062.415"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe shelf to the left of the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004882", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "42446083", "gt_frame_id": "348892.991", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["348850.875", "348847.876", "348839.879", "348837.680", "348833.282", "348880.879"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the right of the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004885", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "42898869", "gt_frame_id": "197835.928", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["197789.430", "197790.629", "197794.428", "197795.627", "197831.830", "197799.226"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004888", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "42899471", "gt_frame_id": "195718.830", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["195724.127", "195715.731", "195714.731", "195749.334", "195744.136", "195748.134"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table under the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004891", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "45260920", "gt_frame_id": "2492.325", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2533.826", "2473.816", "2476.432", "2475.016", "2496.624", "2478.631"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the left of the table, and to the left of the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004894", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "45261182", "gt_frame_id": "2918.832", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["2926.130", "2967.130", "2915.134", "2953.535", "2944.939", "2944.139"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe toilet to the left of the sink.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004897", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "45663149", "gt_frame_id": "63324.901", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["63304.709", "63301.510", "63351.707", "63341.511", "63352.707", "63338.712"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004900", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "45663150", "gt_frame_id": "63537.858", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["63596.152", "63577.759", "63578.359", "63520.149", "63583.557", "63510.952"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004903", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47115543", "gt_frame_id": "15688.622", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["15650.420", "15639.025", "15677.626", "15636.626", "15633.211", "15662.416"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet below the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.5 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004906", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47115550", "gt_frame_id": "15785.917", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["15819.720", "15819.120", "15817.721", "15783.318", "15791.715", "15796.113"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table below the tv_monitor.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004909", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47331063", "gt_frame_id": "129466.153", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["129452.942", "129441.147", "129440.547", "129455.158", "129438.747", "129447.144"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet under the stove.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004912", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47331063", "gt_frame_id": "129465.953", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["129434.549", "129449.743", "129446.744", "129440.347", "129440.947", "129452.742"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe stove on the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004915", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47332886", "gt_frame_id": "7566.066", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["7550.872", "7549.256", "7578.461", "7577.661", "7576.861", "7580.260"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe cabinet to the left of the shelf.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004921", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47333899", "gt_frame_id": "81298.469", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["81253.270", "81310.464", "81309.464", "81292.871", "81303.067", "81284.674"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe sink on the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 0.9 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004924", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "47333934", "gt_frame_id": "85577.763", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["85641.171", "85573.165", "85572.565", "85555.372", "85557.172", "85607.768"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table under the tv_monitor, and to the left of the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.7 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004927", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "48458417", "gt_frame_id": "115293.748", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["115302.744", "115258.946", "115274.756", "115282.153", "115343.744", "115345.143"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe tv_monitor above the cabinet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004930", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "48458430", "gt_frame_id": "116446.343", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["116461.437", "116442.045", "116441.845", "116441.262", "116454.040", "116453.240"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe sink to the left of the toilet.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.8 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004933", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "48458647", "gt_frame_id": "7026.602", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["7032.316", "7022.104", "7050.109", "7050.509", "7045.711", "7016.906"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe table to the right of the chair.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.1 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004936", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "48458652", "gt_frame_id": "7168.410", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["7163.012", "7132.008", "7141.205", "7129.609", "7127.610", "7126.011"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair to the left of the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 2.0 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004939", "dataset": "arkitscenes", "task_type": "obj_loc", "scene_id": "48458667", "gt_frame_id": "8397.059", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["8357.259", "8400.458", "8366.855", "8378.250", "8380.849", "8382.848"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget object:\nThe chair in front of the table.\n\nIdentify the unique object satisfying all stated relations. Position the camera approximately 1.4 meters from the target object's center. Render it clearly near the image center, with every named relation object also visible in the stated spatial configuration.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004942", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42445022", "gt_frame_id": "49084.362", "anchor_frame_id": "49082.863", "shot": 6, "contexts_frame_id": ["49082.863", "49026.268", "49020.271", "49024.469", "49032.466", "49036.065"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the table.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004945", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42445029", "gt_frame_id": "49572.094", "anchor_frame_id": "49590.787", "shot": 6, "contexts_frame_id": ["49590.787", "49529.595", "49568.996", "49567.996", "49539.391", "49541.190"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004948", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42897692", "gt_frame_id": "480636.986", "anchor_frame_id": "480884.583", "shot": 6, "contexts_frame_id": ["480884.583", "480872.988", "480876.186", "480869.789", "480616.594", "480617.794"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the table.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004951", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42897692", "gt_frame_id": "480683.800", "anchor_frame_id": "480739.693", "shot": 6, "contexts_frame_id": ["480739.693", "480737.794", "480723.800", "480722.800", "480652.396", "480620.593"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the shelf.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004954", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42897695", "gt_frame_id": "480356.402", "anchor_frame_id": "480363.199", "shot": 6, "contexts_frame_id": ["480363.199", "480377.793", "480286.198", "480248.796", "480252.195", "480381.992"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the shelf.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004957", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42899461", "gt_frame_id": "195869.734", "anchor_frame_id": "195873.532", "shot": 6, "contexts_frame_id": ["195873.532", "195874.232", "195875.231", "195854.423", "195853.024", "195801.029"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the fireplace.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004960", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "42899471", "gt_frame_id": "195719.130", "anchor_frame_id": "195722.928", "shot": 6, "contexts_frame_id": ["195722.928", "195674.131", "195669.333", "195672.332", "195714.731", "195749.334"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004963", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "45261129", "gt_frame_id": "3357.167", "anchor_frame_id": "3363.764", "shot": 6, "contexts_frame_id": ["3363.764", "3398.367", "3397.967", "3397.567", "3387.971", "3369.562"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the stove.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004966", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "47332885", "gt_frame_id": "7516.169", "anchor_frame_id": "7511.971", "shot": 6, "contexts_frame_id": ["7511.971", "7492.062", "7491.263", "7493.662", "7497.860", "7503.458"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004969", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "47332893", "gt_frame_id": "9274.587", "anchor_frame_id": "9308.290", "shot": 6, "contexts_frame_id": ["9308.290", "9260.893", "9263.692", "9263.492", "9269.889", "9317.286"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004972", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "47334110", "gt_frame_id": "93570.690", "anchor_frame_id": "93611.890", "shot": 6, "contexts_frame_id": ["93611.890", "93613.089", "93590.498", "93535.487", "93587.500", "93584.901"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004975", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "47334110", "gt_frame_id": "93569.590", "anchor_frame_id": "93547.199", "shot": 6, "contexts_frame_id": ["93547.199", "93555.696", "93557.695", "93584.501", "93535.487", "93538.286"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004978", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48018387", "gt_frame_id": "390395.250", "anchor_frame_id": "390415.542", "shot": 6, "contexts_frame_id": ["390415.542", "390384.238", "390411.044", "390384.838", "390386.237", "390414.642"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004981", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48458656", "gt_frame_id": "7654.492", "anchor_frame_id": "7651.793", "shot": 6, "contexts_frame_id": ["7651.793", "7665.388", "7662.589", "7659.990", "7627.787", "7640.781"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the toilet.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004984", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48458732", "gt_frame_id": "139144.787", "anchor_frame_id": "139126.195", "shot": 6, "contexts_frame_id": ["139126.195", "139104.886", "139141.089", "139109.684", "139102.287", "139149.685"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor near the table.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004987", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48458735", "gt_frame_id": "139203.497", "anchor_frame_id": "139210.195", "shot": 6, "contexts_frame_id": ["139210.195", "139172.093", "139241.899", "139229.087", "139229.687", "139250.895"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the stove.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004990", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48458747", "gt_frame_id": "139077.397", "anchor_frame_id": "139052.890", "shot": 6, "contexts_frame_id": ["139052.890", "139050.891", "139063.286", "139039.495", "139073.899", "139032.498"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004993", "dataset": "arkitscenes", "task_type": "loc-re-facing", "scene_id": "48458747", "gt_frame_id": "139077.997", "anchor_frame_id": "139051.591", "shot": 6, "contexts_frame_id": ["139051.591", "139039.096", "139039.695", "139032.498", "139073.682", "139073.899"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-facing instruction:\nTurn to face the tv_monitor.\n\nKeep Image 1's camera center, roll, and field of view fixed. Rotate in place until the target is clearly visible near the image center.\n\nINPUT VIEWS:\nImage 1 is the anchor and defines the starting camera pose; the other images are supporting views of the same scene.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004996", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "41125718", "gt_frame_id": "4767.826", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["4745.518", "4747.118", "4749.717", "4752.732", "4756.730", "4762.728"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the bed to my right, I face the cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00004999", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "41159504", "gt_frame_id": "34711.789", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["34717.786", "34718.186", "34721.385", "34751.189", "34769.981", "34706.791"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the bed, the bathtub is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005002", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "41159571", "gt_frame_id": "34993.224", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["35017.414", "35024.812", "34953.423", "34990.209", "34945.410", "34943.410"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the shelf, with the chair to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005005", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42445981", "gt_frame_id": "347006.530", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["346964.531", "346963.531", "346956.934", "346952.536", "346955.934", "347015.127"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the table, the sofa is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005008", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42446056", "gt_frame_id": "349265.888", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["349291.377", "349314.184", "349281.981", "349278.582", "349251.577", "349248.778"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the cabinet, and the chair remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005011", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897521", "gt_frame_id": "274011.390", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["273968.891", "273960.894", "273962.894", "273962.294", "274008.292", "273956.496"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the stove, with the refrigerator to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005014", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897541", "gt_frame_id": "275436.827", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["275386.731", "275432.928", "275429.130", "275380.333", "275379.734", "275399.326"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the sofa to my left, I face the table.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005017", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897667", "gt_frame_id": "478092.790", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["478022.685", "478096.488", "478097.288", "478087.675", "478082.677", "478064.485"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the stool to my left, I face the table.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005020", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897692", "gt_frame_id": "480683.900", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["480620.593", "480722.800", "480650.197", "480618.394", "480673.987", "480691.996"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the shelf, and the bed is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005023", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897695", "gt_frame_id": "480356.902", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["480282.599", "480286.198", "480377.793", "480361.000", "480307.006", "480240.200"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nWith the bed to my right, I am facing the cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005026", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42897695", "gt_frame_id": "480356.402", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["480382.191", "480380.592", "480307.606", "480376.394", "480286.198", "480240.600"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the bed to my right, I face the shelf.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005029", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42898822", "gt_frame_id": "168214.971", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["168209.573", "168251.573", "168249.973", "168248.774", "168249.574", "168243.376"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the sink to my right, I face the cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005032", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42899620", "gt_frame_id": "228848.672", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["228874.662", "228873.862", "228881.459", "228853.470", "228858.068", "228854.670"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am looking toward the toilet, and the cabinet remains behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005035", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "42899620", "gt_frame_id": "228858.968", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["228873.462", "228883.059", "228848.272", "228853.470", "228854.070", "228854.470"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the sink to my left, I face the toilet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005038", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "44358518", "gt_frame_id": "52854.801", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["52889.203", "52752.011", "52751.811", "52874.410", "52751.211", "52862.414"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the sofa to my right, I face the cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005041", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "45261129", "gt_frame_id": "3395.068", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["3368.162", "3356.767", "3359.566", "3381.757", "3359.966", "3360.965"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the tv_monitor, the table is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005044", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "45662981", "gt_frame_id": "27994.793", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["27948.295", "27949.695", "28016.285", "27946.096", "27967.488", "28034.494"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the sofa to my left, I face the cabinet.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005047", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47331644", "gt_frame_id": "22311.515", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["22305.700", "22333.106", "22301.502", "22335.105", "22343.502", "22328.108"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nNear the chair, I am facing the stool, and I can see the table on the left side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005050", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47331964", "gt_frame_id": "167602.246", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["167553.749", "167502.953", "167555.748", "167492.740", "167598.947", "167597.948"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the sofa, the cabinet is to my right.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005053", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47331990", "gt_frame_id": "170519.754", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["170526.551", "170525.352", "170559.755", "170527.151", "170528.750", "170528.551"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nI am facing the cabinet, with the bed to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005056", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47333898", "gt_frame_id": "81375.271", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["81367.774", "81352.963", "81352.163", "81353.963", "81355.979", "81387.966"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the sink, and the chair is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005059", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47333927", "gt_frame_id": "84025.004", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["84032.401", "83999.798", "84002.613", "84009.610", "84020.206", "84022.005"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nAs I face the stool, the bed is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005062", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47334120", "gt_frame_id": "94394.443", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["94414.052", "94386.047", "94387.846", "94385.447", "94389.045", "94390.445"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the cabinet, and the bed is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005065", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47895341", "gt_frame_id": "96443.625", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["96421.517", "96418.918", "96449.322", "96450.322", "96450.522", "96451.921"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding with the bed to my right, I face the chair.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005068", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "47895355", "gt_frame_id": "98816.928", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["98809.031", "98812.429", "98814.029", "98875.637", "98883.634", "98819.427"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nStanding near the chair, I am facing the sink, and I can see the oven on the left side of the view, and the cabinet on the right side of the view.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005071", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "48018368", "gt_frame_id": "389551.467", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["389580.272", "389580.672", "389540.072", "389539.872", "389539.672", "389539.472"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nFrom here, I face the washer, and the refrigerator is to my left.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}
{"sample_id": "sample_00005074", "dataset": "arkitscenes", "task_type": "view_localization", "scene_id": "48458650", "gt_frame_id": "7063.504", "anchor_frame_id": null, "shot": 6, "contexts_frame_id": ["7095.007", "7098.606", "7099.006", "7070.617", "7079.814", "7068.202"], "instruction": "Generate one photorealistic first-person RGB view of the same static indoor scene.\n\nTARGET CAMERA:\nTarget-camera statement:\nMy view is toward the table, and the cabinet is behind me.\n\nInfer the pose satisfying the full statement and render its forward view. Faced objects and objects assigned to image regions should appear accordingly. Other location cues are not required to be visible, but may appear fully or partly when naturally visible from the target pose and field of view; do not force them into or out of the frame.\n\nINPUT VIEWS:\nAll images are supporting views of the same scene; none is an anchor or target view.\n\nSCENE CONSISTENCY:\nUse the inputs to preserve scene geometry, object identity, appearance, scale, and layout. Render the target perspective, visibility, and occlusion correctly, inferring newly visible content conservatively.\n\nOUTPUT:\nReturn one square, full-frame RGB image without added captions, annotations, borders, collage, or panorama."}