preprocess_summary.json

[
  {
    "episode_id": 4,
    "global_task": "clothes washing",
    "source_file": "videos/observation.images.head/chunk-000/file-002.mp4",
    "episode_start_s_requested": 0.0,
    "episode_end_s": 16.1,
    "decoded_start_frame_index": 0,
    "decoded_start_time_s": 0.0,
    "source_video": {
      "width": 1280,
      "height": 480,
      "fps": 30.0,
      "frames": 5316,
      "duration_s": 177.2
    },
    "condition_image": {
      "path": "condition.png",
      "width": 640,
      "height": 480
    },
    "ground_truth_5s": {
      "path": "ground_truth_5s.mp4",
      "fps": 30.0,
      "frames": 150,
      "duration_s": 5.0,
      "width": 640,
      "height": 480
    },
    "prompt": "A humanoid robot performs the task: \"clothes washing\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
    "model_prompts": {
      "cosmos3_nano": "A humanoid robot performs the task: \"clothes washing\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
      "wan2_1_i2v_14b": "A humanoid robot performs the task: \"clothes washing\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance."
    },
    "inference": {},
    "validation": {
      "source_is_stereo_1280x480": true,
      "left_camera_crop_xywh": [
        0,
        0,
        640,
        480
      ],
      "condition_matches_source_left_crop_exact": true,
      "condition_vs_ground_truth_first_frame_mean_abs_diff_decoded_mp4": 1.6845876736111112,
      "episode_25_start_check_passed": true,
      "no_subtask_annotations_used": true,
      "full_dataset_downloaded": false
    }
  },
  {
    "episode_id": 22,
    "global_task": "picking trash to rubbish bin",
    "source_file": "videos/observation.images.head/chunk-000/file-006.mp4",
    "episode_start_s_requested": 0.0,
    "episode_end_s": 23.833,
    "decoded_start_frame_index": 0,
    "decoded_start_time_s": 0.0,
    "source_video": {
      "width": 1280,
      "height": 480,
      "fps": 30.0,
      "frames": 6927,
      "duration_s": 230.9
    },
    "condition_image": {
      "path": "condition.png",
      "width": 640,
      "height": 480
    },
    "ground_truth_5s": {
      "path": "ground_truth_5s.mp4",
      "fps": 30.0,
      "frames": 150,
      "duration_s": 5.0,
      "width": 640,
      "height": 480
    },
    "prompt": "A humanoid robot performs the task: \"picking trash to rubbish bin\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
    "model_prompts": {
      "cosmos3_nano": "A humanoid robot performs the task: \"picking trash to rubbish bin\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
      "wan2_1_i2v_14b": "A humanoid robot performs the task: \"picking trash to rubbish bin\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance."
    },
    "inference": {},
    "validation": {
      "source_is_stereo_1280x480": true,
      "left_camera_crop_xywh": [
        0,
        0,
        640,
        480
      ],
      "condition_matches_source_left_crop_exact": true,
      "condition_vs_ground_truth_first_frame_mean_abs_diff_decoded_mp4": 1.120470920138889,
      "episode_25_start_check_passed": true,
      "no_subtask_annotations_used": true,
      "full_dataset_downloaded": false
    }
  },
  {
    "episode_id": 25,
    "global_task": "restocking fridge",
    "source_file": "videos/observation.images.head/chunk-000/file-006.mp4",
    "episode_start_s_requested": 102.1,
    "episode_end_s": 230.9,
    "decoded_start_frame_index": 3063,
    "decoded_start_time_s": 102.1,
    "source_video": {
      "width": 1280,
      "height": 480,
      "fps": 30.0,
      "frames": 6927,
      "duration_s": 230.9
    },
    "condition_image": {
      "path": "condition.png",
      "width": 640,
      "height": 480
    },
    "ground_truth_5s": {
      "path": "ground_truth_5s.mp4",
      "fps": 30.0,
      "frames": 150,
      "duration_s": 5.0,
      "width": 640,
      "height": 480
    },
    "prompt": "A humanoid robot performs the task: \"restocking fridge\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
    "model_prompts": {
      "cosmos3_nano": "A humanoid robot performs the task: \"restocking fridge\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
      "wan2_1_i2v_14b": "A humanoid robot performs the task: \"restocking fridge\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance."
    },
    "inference": {},
    "validation": {
      "source_is_stereo_1280x480": true,
      "left_camera_crop_xywh": [
        0,
        0,
        640,
        480
      ],
      "condition_matches_source_left_crop_exact": true,
      "condition_vs_ground_truth_first_frame_mean_abs_diff_decoded_mp4": 1.445615234375,
      "episode_25_start_check_passed": true,
      "no_subtask_annotations_used": true,
      "full_dataset_downloaded": false
    }
  },
  {
    "episode_id": 22983,
    "global_task": "building children table",
    "source_file": "videos/observation.images.head/chunk-005/file-909.mp4",
    "episode_start_s_requested": 0.0,
    "episode_end_s": 831.9,
    "decoded_start_frame_index": 0,
    "decoded_start_time_s": 0.0,
    "source_video": {
      "width": 1280,
      "height": 480,
      "fps": 30.0,
      "frames": 24957,
      "duration_s": 831.9
    },
    "condition_image": {
      "path": "condition.png",
      "width": 640,
      "height": 480
    },
    "ground_truth_5s": {
      "path": "ground_truth_5s.mp4",
      "fps": 30.0,
      "frames": 150,
      "duration_s": 5.0,
      "width": 640,
      "height": 480
    },
    "prompt": "A humanoid robot performs the task: \"building children table\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
    "model_prompts": {
      "cosmos3_nano": "A humanoid robot performs the task: \"building children table\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance.",
      "wan2_1_i2v_14b": "A humanoid robot performs the task: \"building children table\". Continue the scene from the provided first-person robot-camera image and generate the physically plausible future video showing the robot beginning to execute this task. Preserve the same environment, objects, robot embodiment, camera viewpoint, lighting, and visual appearance."
    },
    "inference": {},
    "validation": {
      "source_is_stereo_1280x480": true,
      "left_camera_crop_xywh": [
        0,
        0,
        640,
        480
      ],
      "condition_matches_source_left_crop_exact": true,
      "condition_vs_ground_truth_first_frame_mean_abs_diff_decoded_mp4": 1.0727235243055555,
      "episode_25_start_check_passed": true,
      "no_subtask_annotations_used": true,
      "full_dataset_downloaded": false
    }
  }
]