Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
17
17
sample_index
int64
1
1.2k
level
stringclasses
3 values
conversion_level
stringclasses
3 values
layout_family
stringclasses
3 values
camera_view
stringclasses
2 values
camera_viewpoint
stringclasses
2 values
prompt_mode
stringclasses
1 value
scene_slice
stringclasses
4 values
scene_context
stringclasses
4 values
FoR_prompt
stringlengths
250
509
Cam_prompt
stringlengths
241
471
original_FoR_prompt
stringlengths
250
509
original_Cam_prompt
stringlengths
241
471
layout
dict
layout_result
dict
layout_signature
dict
construction
unknown
objects
unknown
target_anchor_relation
stringclasses
4 values
target_image_relation
stringclasses
6 values
relationship_family
stringclasses
3 values
relation
stringclasses
4 values
object_orientation
stringclasses
6 values
orientation_rule
stringclasses
4 values
anchor_category
stringclasses
3 values
benchmark
stringclasses
1 value
split
stringclasses
1 value
license_note
stringclasses
1 value
FoR-T2I-v1-000001
1
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A woman is present, facing image-right. A camel is present. The camel faces the woman. From woman's perspective, camel is to its left. An airplane is present. The airplane is behind camel, opposite camel's own facing direction. All objects are shown at a similar visual size, with small clear ...
Create an eye-level scene. A woman is present, facing image-right. A camel is present. The camel is facing the viewer. From the viewer's perspective, camel is in the background relative to woman. An airplane is present. From the viewer's perspective, airplane appears farther away than camel. All objects are shown at a ...
Create an eye-level scene. A woman is present, facing image-right. A camel is present. The camel faces the woman. From woman's perspective, camel is to its left. An airplane is present. The airplane is behind camel, opposite camel's own facing direction. All objects are shown at a similar visual size, with small clear ...
Create an eye-level scene. A woman is present, facing image-right. A camel is present. The camel is facing the viewer. From the viewer's perspective, camel is in the background relative to woman. An airplane is present. From the viewer's perspective, airplane appears farther away than camel. All objects are shown at a ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 128, "y": 240.64, "label": "woman", "direction": "facing_image_right", "is_an...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "woman", "x": 128, "y": 240.64, "direction": "facing_image_right", "is_an...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "woman", "x": 128, "y": 240.64, "direction": "facing_image_right", "is_an...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "airplane", "orientation_source_anchor": "woman", "orientation_rule": "faces_source", "orientation_source_direction": "facing_image_right", "resolved_anchor_direction": "facing_viewer", "second_anchor": ...
{ "anchor": "camel", "orientation_source_anchor": "woman", "target": "airplane", "targets": [ "airplane" ], "distractors": [], "second_anchor": "woman" }
anchor_back
background
depth_foreground_background
anchor_back
facing_viewer
faces_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000002
2
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A van is present, facing image-top. A cell phone is present. The cell phone is on van's own right side, from van's perspective. A horse is present. A horse is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gap...
Create a top-down map. A van is present, facing image-top. A cell phone is present. Viewed directly, cell phone appears to the right of van. A horse is present. A horse is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between ...
Create a top-down map. A van is present, facing image-top. A cell phone is present. The cell phone is on van's own right side, from van's perspective. A horse is present. A horse is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gap...
Create a top-down map. A van is present, facing image-top. A cell phone is present. Viewed directly, cell phone appears to the right of van. A horse is present. A horse is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "van", "direction": "facing_image_top", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "van", "x": 128, "y": 128, "direction": "facing_image_top", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "van", "x": 128, "y": 128, "direction": "facing_image_top", "is_anchor": true }, ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_right" ], "image_level_target_id": "target_2", "for_target_label": "cell phone", "image_level_target_label": "horse", "image_level_position": "image_left", "image_level_constraint": "viewer/image-frame position indepen...
{ "anchor": "van", "orientation_source_anchor": null, "target": "cell phone", "targets": [ "cell phone", "horse" ], "distractors": [] }
anchor_right
image_right
screen_left_right
anchor_right
facing_image_top
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000003
3
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A red panda is present, facing image-right. A polar bear is present. The polar bear faces away from the red panda. From red panda's perspective, polar bear is directly in front of it. A sheep is present. The sheep is on polar bear's own right side, from polar bear's perspective. All objects a...
Create an eye-level scene. A red panda is present, facing image-right. A polar bear is present. The polar bear is facing image-right. The polar bear is to the right of red panda from the viewer's perspective. A sheep is present. From the viewer's perspective, sheep appears closer than polar bear. All objects are shown ...
Create an eye-level scene. A red panda is present, facing image-right. A polar bear is present. The polar bear faces away from the red panda. From red panda's perspective, polar bear is directly in front of it. A sheep is present. The sheep is on polar bear's own right side, from polar bear's perspective. All objects a...
Create an eye-level scene. A red panda is present, facing image-right. A polar bear is present. The polar bear is facing image-right. The polar bear is to the right of red panda from the viewer's perspective. A sheep is present. From the viewer's perspective, sheep appears closer than polar bear. All objects are shown ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "red panda", "direction": "facing_image_right", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "red panda", "x": 15.36, "y": 128, "direction": "facing_image_right", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "red panda", "x": 15.36, "y": 128, "direction": "facing_image_right", "is...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "sheep", "orientation_source_anchor": "red panda", "orientation_rule": "faces_away_from_source", "orientation_source_direction": "facing_image_right", "resolved_anchor_direction": "facing_image_right", ...
{ "anchor": "polar bear", "orientation_source_anchor": "red panda", "target": "sheep", "targets": [ "sheep" ], "distractors": [], "second_anchor": "red panda" }
anchor_right
foreground
depth_foreground_background
anchor_right
facing_image_right
faces_away_from_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000004
4
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A suv is present, facing image-left. A cat is present. The cat is on suv's own right side, from suv's perspective. A bird is present. A bird is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between them ...
Create a top-down map. A suv is present, facing image-left. A cat is present. The cat is above suv from the viewer's perspective. A bird is present. A bird is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between them so none ...
Create a top-down map. A suv is present, facing image-left. A cat is present. The cat is on suv's own right side, from suv's perspective. A bird is present. A bird is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between them ...
Create a top-down map. A suv is present, facing image-left. A cat is present. The cat is above suv from the viewer's perspective. A bird is present. A bird is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps between them so none ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "suv", "direction": "facing_image_left", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "suv", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "suv", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }, ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_right" ], "image_level_target_id": "target_2", "for_target_label": "cat", "image_level_target_label": "bird", "image_level_position": "image_left", "image_level_constraint": "viewer/image-frame position independent of ...
{ "anchor": "suv", "orientation_source_anchor": null, "target": "cat", "targets": [ "cat", "bird" ], "distractors": [] }
anchor_right
image_top
other
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000005
5
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A high heels is present, facing image-left. A cow is present. The cow is on high heels's own right side, from high heels's perspective. A chair is present. A chair is placed near the bottom of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cl...
Create a top-down map. A high heels is present, facing image-left. A cow is present. The cow is above high heels from the viewer's perspective. A chair is present. A chair is placed near the bottom of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps betwee...
Create a top-down map. A high heels is present, facing image-left. A cow is present. The cow is on high heels's own right side, from high heels's perspective. A chair is present. A chair is placed near the bottom of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cl...
Create a top-down map. A high heels is present, facing image-left. A cow is present. The cow is above high heels from the viewer's perspective. A chair is present. A chair is placed near the bottom of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps betwee...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "high heels", "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "high heels", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "high heels", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_right" ], "image_level_target_id": "target_2", "for_target_label": "cow", "image_level_target_label": "chair", "image_level_position": "image_bottom", "image_level_constraint": "viewer/image-frame position independent ...
{ "anchor": "high heels", "orientation_source_anchor": null, "target": "cow", "targets": [ "cow", "chair" ], "distractors": [] }
anchor_right
image_top
other
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000006
6
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A swan is present, facing image-left. A kite is present. The kite sits on the forward side of swan, using swan's own orientation. A hair drier is present. A hair drier is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with sm...
Create a top-down map. A swan is present, facing image-left. A kite is present. From the viewer's perspective, kite is positioned left of swan. A hair drier is present. A hair drier is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear...
Create a top-down map. A swan is present, facing image-left. A kite is present. The kite sits on the forward side of swan, using swan's own orientation. A hair drier is present. A hair drier is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with sm...
Create a top-down map. A swan is present, facing image-left. A kite is present. From the viewer's perspective, kite is positioned left of swan. A hair drier is present. A hair drier is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "swan", "direction": "facing_image_left", "is_anchor": true },...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "swan", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true },...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "swan", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true },...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_front" ], "image_level_target_id": "target_2", "for_target_label": "kite", "image_level_target_label": "hair drier", "image_level_position": "image_right", "image_level_constraint": "viewer/image-frame position indepen...
{ "anchor": "swan", "orientation_source_anchor": null, "target": "kite", "targets": [ "kite", "hair drier" ], "distractors": [] }
anchor_front
image_left
screen_left_right
anchor_front
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000007
7
L2
layout_multi_relation
multi_relation
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. An oven is present, facing image-left. A tv remote is present. The tv remote is behind oven, opposite oven's own facing direction. A train is present. A train is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear ...
Create an eye-level scene. An oven is present, facing image-left. A tv remote is present. From the viewer's perspective, tv remote is positioned right of oven. A train is present. A train is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clea...
Create an eye-level scene. An oven is present, facing image-left. A tv remote is present. The tv remote is behind oven, opposite oven's own facing direction. A train is present. A train is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear ...
Create an eye-level scene. An oven is present, facing image-left. A tv remote is present. From the viewer's perspective, tv remote is positioned right of oven. A train is present. A train is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clea...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "oven", "direction": "facing_image_left", "is_anchor": true }...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "oven", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "oven", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_back" ], "image_level_target_id": "target_2", "for_target_label": "tv remote", "image_level_target_label": "train", "image_level_position": "image_top", "image_level_constraint": "viewer/image-frame position independen...
{ "anchor": "oven", "orientation_source_anchor": null, "target": "tv remote", "targets": [ "tv remote", "train" ], "distractors": [] }
anchor_back
image_right
screen_left_right
anchor_back
facing_image_left
null
stable_side_axis_anchor
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000008
8
L1
layout_single_anchor
single_anchor
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A binoculars is present, facing image-right. A boat is present. The boat is behind binoculars, opposite binoculars's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A binoculars is present, facing image-right. A boat is present. The boat is to the left of binoculars from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A binoculars is present, facing image-right. A boat is present. The boat is behind binoculars, opposite binoculars's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A binoculars is present, facing image-right. A boat is present. The boat is to the left of binoculars from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "binoculars", "direction": "facing_image_right", "is_anchor": tru...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "binoculars", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": tru...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "binoculars", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": tru...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "boat" }
{ "anchor": "binoculars", "orientation_source_anchor": null, "target": "boat", "targets": [ "boat" ], "distractors": [] }
anchor_back
image_left
screen_left_right
anchor_back
facing_image_right
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000009
9
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
character_blocking
a clear character blocking scene with the two subjects separated
Create a top-down map. A cat is present, facing image-top. An elephant is present. The elephant is behind cat, opposite cat's own facing direction. A zebra is present. A zebra is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps b...
Create a top-down map. A cat is present, facing image-top. An elephant is present. From the viewer's perspective, elephant is positioned below cat. A zebra is present. A zebra is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps b...
Create a top-down map. A cat is present, facing image-top. An elephant is present. The elephant is behind cat, opposite cat's own facing direction. A zebra is present. A zebra is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps b...
Create a top-down map. A cat is present, facing image-top. An elephant is present. From the viewer's perspective, elephant is positioned below cat. A zebra is present. A zebra is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps b...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "cat", "direction": "facing_image_top", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "cat", "x": 128, "y": 128, "direction": "facing_image_top", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "cat", "x": 128, "y": 128, "direction": "facing_image_top", "is_anchor": true }, ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_back" ], "image_level_target_id": "target_2", "for_target_label": "elephant", "image_level_target_label": "zebra", "image_level_position": "image_left", "image_level_constraint": "viewer/image-frame position independen...
{ "anchor": "cat", "orientation_source_anchor": null, "target": "elephant", "targets": [ "elephant", "zebra" ], "distractors": [] }
anchor_back
image_bottom
other
anchor_back
facing_image_top
null
movable_or_deformable
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000010
10
L1
layout_single_anchor
single_anchor
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A raccoon is present, facing image-left. A bed is present. The bed sits on the forward side of raccoon, using raccoon's own orientation. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A raccoon is present, facing image-left. A bed is present. From the viewer's perspective, bed is positioned left of raccoon. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A raccoon is present, facing image-left. A bed is present. The bed sits on the forward side of raccoon, using raccoon's own orientation. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A raccoon is present, facing image-left. A bed is present. From the viewer's perspective, bed is positioned left of raccoon. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "raccoon", "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "raccoon", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "raccoon", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_front" ], "for_target_label": "bed" }
{ "anchor": "raccoon", "orientation_source_anchor": null, "target": "bed", "targets": [ "bed" ], "distractors": [] }
anchor_front
image_left
screen_left_right
anchor_front
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000011
11
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
character_blocking
a clear character blocking scene with the two subjects separated
Create an eye-level scene. A tank is present, facing away from the viewer. A fox is present. The fox faces the opposite direction from the tank. The fox sits on the right-hand side of tank, using tank's own orientation. A fork is present. The fork sits on the left-hand side of fox, using fox's own orientation. All obje...
Create an eye-level scene. A tank is present, facing away from the viewer. A fox is present. The fox is facing the viewer. The fox is to the right of tank from the viewer's perspective. A fork is present. From the viewer's perspective, fork is positioned right of fox. All objects are shown at a similar visual size, wit...
Create an eye-level scene. A tank is present, facing away from the viewer. A fox is present. The fox faces the opposite direction from the tank. The fox sits on the right-hand side of tank, using tank's own orientation. A fork is present. The fork sits on the left-hand side of fox, using fox's own orientation. All obje...
Create an eye-level scene. A tank is present, facing away from the viewer. A fox is present. The fox is facing the viewer. The fox is to the right of tank from the viewer's perspective. A fork is present. From the viewer's perspective, fork is positioned right of fox. All objects are shown at a similar visual size, wit...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "tank", "direction": "facing_away_from_viewer", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "tank", "x": 15.36, "y": 128, "direction": "facing_away_from_viewer", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "tank", "x": 15.36, "y": 128, "direction": "facing_away_from_viewer", "is...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_left" ], "for_target_label": "fork", "orientation_source_anchor": "tank", "orientation_rule": "opposite_direction", "orientation_source_direction": "facing_away_from_viewer", "resolved_anchor_direction": "facing_viewer", "second_anc...
{ "anchor": "fox", "orientation_source_anchor": "tank", "target": "fork", "targets": [ "fork" ], "distractors": [], "second_anchor": "tank" }
anchor_left
image_right
screen_left_right
anchor_left
facing_viewer
opposite_direction
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000012
12
L1
layout_single_anchor
single_anchor
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A men is present, facing image-left. A car is present. The car is on men's own right side, from men's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A men is present, facing image-left. A car is present. The car is above men from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A men is present, facing image-left. A car is present. The car is on men's own right side, from men's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A men is present, facing image-left. A car is present. The car is above men from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "men", "direction": "facing_image_left", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "men", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }, ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "men", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true }, ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "car" }
{ "anchor": "men", "orientation_source_anchor": null, "target": "car", "targets": [ "car" ], "distractors": [] }
anchor_right
image_top
other
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000013
13
L1
layout_single_anchor
single_anchor
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A panda is present, facing image-bottom. A backpack is present. The backpack is behind panda, opposite panda's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A panda is present, facing image-bottom. A backpack is present. The backpack is above panda from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A panda is present, facing image-bottom. A backpack is present. The backpack is behind panda, opposite panda's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A panda is present, facing image-bottom. A backpack is present. The backpack is above panda from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "panda", "direction": "facing_image_bottom", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "panda", "x": 128, "y": 128, "direction": "facing_image_bottom", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "panda", "x": 128, "y": 128, "direction": "facing_image_bottom", "is_anchor": true ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "backpack" }
{ "anchor": "panda", "orientation_source_anchor": null, "target": "backpack", "targets": [ "backpack" ], "distractors": [] }
anchor_back
image_top
other
anchor_back
facing_image_bottom
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000014
14
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A wheelchair is present, facing image-left. A crane is present. The crane faces the same direction as the wheelchair. The crane sits on the rear side of wheelchair, using wheelchair's own orientation. A computer keyboard is present. The computer keyboard is behind crane, opposite crane's own ...
Create an eye-level scene. A wheelchair is present, facing image-left. A crane is present. The crane is facing image-left. The crane is to the right of wheelchair from the viewer's perspective. A computer keyboard is present. From the viewer's perspective, computer keyboard is positioned right of crane. All objects are...
Create an eye-level scene. A wheelchair is present, facing image-left. A crane is present. The crane faces the same direction as the wheelchair. The crane sits on the rear side of wheelchair, using wheelchair's own orientation. A computer keyboard is present. The computer keyboard is behind crane, opposite crane's own ...
Create an eye-level scene. A wheelchair is present, facing image-left. A crane is present. The crane is facing image-left. The crane is to the right of wheelchair from the viewer's perspective. A computer keyboard is present. From the viewer's perspective, computer keyboard is positioned right of crane. All objects are...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "wheelchair", "direction": "facing_image_left", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "wheelchair", "x": 15.36, "y": 128, "direction": "facing_image_left", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "wheelchair", "x": 15.36, "y": 128, "direction": "facing_image_left", "is...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "computer keyboard", "orientation_source_anchor": "wheelchair", "orientation_rule": "same_direction", "orientation_source_direction": "facing_image_left", "resolved_anchor_direction": "facing_image_left", ...
{ "anchor": "crane", "orientation_source_anchor": "wheelchair", "target": "computer keyboard", "targets": [ "computer keyboard" ], "distractors": [], "second_anchor": "wheelchair" }
anchor_back
image_right
screen_left_right
anchor_back
facing_image_left
same_direction
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000015
15
L1
layout_single_anchor
single_anchor
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A blue jay is present, facing image-left. A book is present. The book is on blue jay's own right side, from blue jay's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A blue jay is present, facing image-left. A book is present. From the viewer's perspective, book appears farther away than blue jay. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A blue jay is present, facing image-left. A book is present. The book is on blue jay's own right side, from blue jay's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create an eye-level scene. A blue jay is present, facing image-left. A book is present. From the viewer's perspective, book appears farther away than blue jay. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "blue jay", "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "blue jay", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "blue jay", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "book" }
{ "anchor": "blue jay", "orientation_source_anchor": null, "target": "book", "targets": [ "book" ], "distractors": [] }
anchor_right
background
depth_foreground_background
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000016
16
L3
layout_orientation_link
orientation_link
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A bear is present, facing image-bottom. A cheetah is present. The cheetah faces the same direction as the bear. From bear's perspective, cheetah is to its left. A knife is present. The knife sits on the left-hand side of cheetah, using cheetah's own orientation. All objects are shown at a similar...
Create a top-down map. A bear is present, facing image-bottom. A cheetah is present. The cheetah is facing image-bottom. From the viewer's perspective, cheetah is positioned right of bear. A knife is present. Viewed directly, knife appears to the right of cheetah. All objects are shown at a similar visual size, with sm...
Create a top-down map. A bear is present, facing image-bottom. A cheetah is present. The cheetah faces the same direction as the bear. From bear's perspective, cheetah is to its left. A knife is present. The knife sits on the left-hand side of cheetah, using cheetah's own orientation. All objects are shown at a similar...
Create a top-down map. A bear is present, facing image-bottom. A cheetah is present. The cheetah is facing image-bottom. From the viewer's perspective, cheetah is positioned right of bear. A knife is present. Viewed directly, knife appears to the right of cheetah. All objects are shown at a similar visual size, with sm...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "bear", "direction": "facing_image_bottom", "is_anch...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "bear", "x": 15.36, "y": 128, "direction": "facing_image_bottom", "is_anch...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "bear", "x": 15.36, "y": 128, "direction": "facing_image_bottom", "is_anch...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_left" ], "for_target_label": "knife", "orientation_source_anchor": "bear", "orientation_rule": "same_direction", "orientation_source_direction": "facing_image_bottom", "resolved_anchor_direction": "facing_image_bottom", "second_anch...
{ "anchor": "cheetah", "orientation_source_anchor": "bear", "target": "knife", "targets": [ "knife" ], "distractors": [], "second_anchor": "bear" }
anchor_left
image_right
screen_left_right
anchor_left
facing_image_bottom
same_direction
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000017
17
L1
layout_single_anchor
single_anchor
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A glasses is present, facing image-left. A computer mouse is present. The computer mouse is on glasses's own right side, from glasses's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A glasses is present, facing image-left. A computer mouse is present. The computer mouse is above glasses from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A glasses is present, facing image-left. A computer mouse is present. The computer mouse is on glasses's own right side, from glasses's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. A glasses is present, facing image-left. A computer mouse is present. The computer mouse is above glasses from the viewer's perspective. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "glasses", "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "glasses", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "glasses", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "computer mouse" }
{ "anchor": "glasses", "orientation_source_anchor": null, "target": "computer mouse", "targets": [ "computer mouse" ], "distractors": [] }
anchor_right
image_top
other
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000018
18
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A shark is present, facing image-right. A birds is present. The birds faces the shark. The birds sits on the right-hand side of shark, using shark's own orientation. A person is present. The person is on birds's own right side, from birds's perspective. All objects are shown at a similar visu...
Create an eye-level scene. A shark is present, facing image-right. A birds is present. The birds is facing away from the viewer. From the viewer's perspective, birds is in the foreground relative to shark. A person is present. From the viewer's perspective, person is positioned right of birds. All objects are shown at ...
Create an eye-level scene. A shark is present, facing image-right. A birds is present. The birds faces the shark. The birds sits on the right-hand side of shark, using shark's own orientation. A person is present. The person is on birds's own right side, from birds's perspective. All objects are shown at a similar visu...
Create an eye-level scene. A shark is present, facing image-right. A birds is present. The birds is facing away from the viewer. From the viewer's perspective, birds is in the foreground relative to shark. A person is present. From the viewer's perspective, person is positioned right of birds. All objects are shown at ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 128, "y": 15.36, "label": "shark", "direction": "facing_image_right", "is_anc...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "shark", "x": 128, "y": 15.36, "direction": "facing_image_right", "is_anc...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "shark", "x": 128, "y": 15.36, "direction": "facing_image_right", "is_anc...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "person", "orientation_source_anchor": "shark", "orientation_rule": "faces_source", "orientation_source_direction": "facing_image_right", "resolved_anchor_direction": "facing_away_from_viewer", "second_...
{ "anchor": "birds", "orientation_source_anchor": "shark", "target": "person", "targets": [ "person" ], "distractors": [], "second_anchor": "shark" }
anchor_right
image_right
screen_left_right
anchor_right
facing_away_from_viewer
faces_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000019
19
L3
layout_orientation_link
orientation_link
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A crown is present, facing image-left. A dragonfly is present. The dragonfly faces the opposite direction from the crown. The dragonfly sits on the rear side of crown, using crown's own orientation. A scissors is present. The scissors is on dragonfly's own right side, from dragonfly's perspective...
Create a top-down map. A crown is present, facing image-left. A dragonfly is present. The dragonfly is facing image-right. From the viewer's perspective, dragonfly is positioned right of crown. A scissors is present. From the viewer's perspective, scissors is positioned below dragonfly. All objects are shown at a simil...
Create a top-down map. A crown is present, facing image-left. A dragonfly is present. The dragonfly faces the opposite direction from the crown. The dragonfly sits on the rear side of crown, using crown's own orientation. A scissors is present. The scissors is on dragonfly's own right side, from dragonfly's perspective...
Create a top-down map. A crown is present, facing image-left. A dragonfly is present. The dragonfly is facing image-right. From the viewer's perspective, dragonfly is positioned right of crown. A scissors is present. From the viewer's perspective, scissors is positioned below dragonfly. All objects are shown at a simil...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "crown", "direction": "facing_image_left", "is_ancho...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "crown", "x": 15.36, "y": 128, "direction": "facing_image_left", "is_ancho...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "crown", "x": 15.36, "y": 128, "direction": "facing_image_left", "is_ancho...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_right" ], "for_target_label": "scissors", "orientation_source_anchor": "crown", "orientation_rule": "opposite_direction", "orientation_source_direction": "facing_image_left", "resolved_anchor_direction": "facing_image_right", "secon...
{ "anchor": "dragonfly", "orientation_source_anchor": "crown", "target": "scissors", "targets": [ "scissors" ], "distractors": [], "second_anchor": "crown" }
anchor_right
image_bottom
other
anchor_right
facing_image_right
opposite_direction
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000020
20
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A baseball player is present, facing image-left. A bear is present. The bear is on baseball player's own right side, from baseball player's perspective. A truck is present. A truck is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual ...
Create a top-down map. A baseball player is present, facing image-left. A bear is present. The bear is above baseball player from the viewer's perspective. A truck is present. A truck is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cle...
Create a top-down map. A baseball player is present, facing image-left. A bear is present. The bear is on baseball player's own right side, from baseball player's perspective. A truck is present. A truck is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual ...
Create a top-down map. A baseball player is present, facing image-left. A bear is present. The bear is above baseball player from the viewer's perspective. A truck is present. A truck is placed on the right side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cle...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "baseball player", "direction": "facing_image_left", "is_anchor": ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "baseball player", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "baseball player", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_right" ], "image_level_target_id": "target_2", "for_target_label": "bear", "image_level_target_label": "truck", "image_level_position": "image_right", "image_level_constraint": "viewer/image-frame position independent ...
{ "anchor": "baseball player", "orientation_source_anchor": null, "target": "bear", "targets": [ "bear", "truck" ], "distractors": [] }
anchor_right
image_top
other
anchor_right
facing_image_left
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000021
21
L2
layout_multi_relation
multi_relation
eye_level
eye_level
natural
character_blocking
a clear character blocking scene with the two subjects separated
Create an eye-level scene. A chicken is present, facing image-left. A laptop is present. The laptop sits on the left-hand side of chicken, using chicken's own orientation. A tv is present. A tv is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with smal...
Create an eye-level scene. A chicken is present, facing image-left. A laptop is present. From the viewer's perspective, laptop appears closer than chicken. A tv is present. A tv is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps bet...
Create an eye-level scene. A chicken is present, facing image-left. A laptop is present. The laptop sits on the left-hand side of chicken, using chicken's own orientation. A tv is present. A tv is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with smal...
Create an eye-level scene. A chicken is present, facing image-left. A laptop is present. From the viewer's perspective, laptop appears closer than chicken. A tv is present. A tv is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear gaps bet...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "chicken", "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "chicken", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "chicken", "x": 128, "y": 128, "direction": "facing_image_left", "is_anchor": true ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_left" ], "image_level_target_id": "target_2", "for_target_label": "laptop", "image_level_target_label": "tv", "image_level_position": "image_top", "image_level_constraint": "viewer/image-frame position independent of t...
{ "anchor": "chicken", "orientation_source_anchor": null, "target": "laptop", "targets": [ "laptop", "tv" ], "distractors": [] }
anchor_left
foreground
depth_foreground_background
anchor_left
facing_image_left
null
movable_or_deformable
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000022
22
L3
layout_orientation_link
orientation_link
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A small vehicle is present, facing image-bottom. An alpaca is present. The alpaca faces the same direction as the small vehicle. From small vehicle's perspective, alpaca is to its left. A bicycle is present. The bicycle sits on the forward side of alpaca, using alpaca's own orientation. All objec...
Create a top-down map. A small vehicle is present, facing image-bottom. An alpaca is present. The alpaca is facing image-bottom. From the viewer's perspective, alpaca is positioned right of small vehicle. A bicycle is present. From the viewer's perspective, bicycle is positioned below alpaca. All objects are shown at a...
Create a top-down map. A small vehicle is present, facing image-bottom. An alpaca is present. The alpaca faces the same direction as the small vehicle. From small vehicle's perspective, alpaca is to its left. A bicycle is present. The bicycle sits on the forward side of alpaca, using alpaca's own orientation. All objec...
Create a top-down map. A small vehicle is present, facing image-bottom. An alpaca is present. The alpaca is facing image-bottom. From the viewer's perspective, alpaca is positioned right of small vehicle. A bicycle is present. From the viewer's perspective, bicycle is positioned below alpaca. All objects are shown at a...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "small vehicle", "direction": "facing_image_bottom", ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "small vehicle", "x": 15.36, "y": 128, "direction": "facing_image_bottom", ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "small vehicle", "x": 15.36, "y": 128, "direction": "facing_image_bottom", ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_front" ], "for_target_label": "bicycle", "orientation_source_anchor": "small vehicle", "orientation_rule": "same_direction", "orientation_source_direction": "facing_image_bottom", "resolved_anchor_direction": "facing_image_bottom", ...
{ "anchor": "alpaca", "orientation_source_anchor": "small vehicle", "target": "bicycle", "targets": [ "bicycle" ], "distractors": [], "second_anchor": "small vehicle" }
anchor_front
image_bottom
other
anchor_front
facing_image_bottom
same_direction
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000023
23
L2
layout_multi_relation
multi_relation
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A seahorse is present, facing image-right. An oven is present. The oven sits on the forward side of seahorse, using seahorse's own orientation. A teddy bear is present. A teddy bear is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual ...
Create a top-down map. A seahorse is present, facing image-right. An oven is present. Viewed directly, oven appears to the right of seahorse. A teddy bear is present. A teddy bear is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear ga...
Create a top-down map. A seahorse is present, facing image-right. An oven is present. The oven sits on the forward side of seahorse, using seahorse's own orientation. A teddy bear is present. A teddy bear is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual ...
Create a top-down map. A seahorse is present, facing image-right. An oven is present. Viewed directly, oven appears to the right of seahorse. A teddy bear is present. A teddy bear is placed on the left side of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear ga...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "seahorse", "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "seahorse", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "seahorse", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_front" ], "image_level_target_id": "target_2", "for_target_label": "oven", "image_level_target_label": "teddy bear", "image_level_position": "image_left", "image_level_constraint": "viewer/image-frame position independ...
{ "anchor": "seahorse", "orientation_source_anchor": null, "target": "oven", "targets": [ "oven", "teddy bear" ], "distractors": [] }
anchor_front
image_right
screen_left_right
anchor_front
facing_image_right
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000024
24
L3
layout_orientation_link
orientation_link
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. A tortoise is present, facing image-left. A coffee machine is present. The coffee machine faces away from the tortoise. The coffee machine sits on the rear side of tortoise, using tortoise's own orientation. A clock is present. The clock sits on the left-hand side of coffee machine, using coffee ...
Create a top-down map. A tortoise is present, facing image-left. A coffee machine is present. The coffee machine is facing image-right. From the viewer's perspective, coffee machine is positioned right of tortoise. A clock is present. The clock is above coffee machine from the viewer's perspective. All objects are show...
Create a top-down map. A tortoise is present, facing image-left. A coffee machine is present. The coffee machine faces away from the tortoise. The coffee machine sits on the rear side of tortoise, using tortoise's own orientation. A clock is present. The clock sits on the left-hand side of coffee machine, using coffee ...
Create a top-down map. A tortoise is present, facing image-left. A coffee machine is present. The coffee machine is facing image-right. From the viewer's perspective, coffee machine is positioned right of tortoise. A clock is present. The clock is above coffee machine from the viewer's perspective. All objects are show...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "tortoise", "direction": "facing_image_left", "is_an...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "tortoise", "x": 15.36, "y": 128, "direction": "facing_image_left", "is_an...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "tortoise", "x": 15.36, "y": 128, "direction": "facing_image_left", "is_an...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_left" ], "for_target_label": "clock", "orientation_source_anchor": "tortoise", "orientation_rule": "faces_away_from_source", "orientation_source_direction": "facing_image_left", "resolved_anchor_direction": "facing_image_right", "se...
{ "anchor": "coffee machine", "orientation_source_anchor": "tortoise", "target": "clock", "targets": [ "clock" ], "distractors": [], "second_anchor": "tortoise" }
anchor_left
image_top
other
anchor_left
facing_image_right
faces_away_from_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000025
25
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A goldfish is present, facing image-right. A raven is present. The raven faces the goldfish. The raven sits on the rear side of goldfish, using goldfish's own orientation. A bus is present. The bus is behind raven, opposite raven's own facing direction. All objects are shown at a similar visu...
Create an eye-level scene. A goldfish is present, facing image-right. A raven is present. The raven is facing image-right. Viewed directly, raven appears to the left of goldfish. A bus is present. The bus is to the left of raven from the viewer's perspective. All objects are shown at a similar visual size, with small c...
Create an eye-level scene. A goldfish is present, facing image-right. A raven is present. The raven faces the goldfish. The raven sits on the rear side of goldfish, using goldfish's own orientation. A bus is present. The bus is behind raven, opposite raven's own facing direction. All objects are shown at a similar visu...
Create an eye-level scene. A goldfish is present, facing image-right. A raven is present. The raven is facing image-right. Viewed directly, raven appears to the left of goldfish. A bus is present. The bus is to the left of raven from the viewer's perspective. All objects are shown at a similar visual size, with small c...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 240.64, "y": 128, "label": "goldfish", "direction": "facing_image_right", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "goldfish", "x": 240.64, "y": 128, "direction": "facing_image_right", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "goldfish", "x": 240.64, "y": 128, "direction": "facing_image_right", "is...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "bus", "orientation_source_anchor": "goldfish", "orientation_rule": "faces_source", "orientation_source_direction": "facing_image_right", "resolved_anchor_direction": "facing_image_right", "second_anchor...
{ "anchor": "raven", "orientation_source_anchor": "goldfish", "target": "bus", "targets": [ "bus" ], "distractors": [], "second_anchor": "goldfish" }
anchor_back
image_left
screen_left_right
anchor_back
facing_image_right
faces_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000026
26
L2
layout_multi_relation
multi_relation
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A rickshaw is present, facing image-right. A spoon is present. The spoon is on rickshaw's own right side, from rickshaw's perspective. A woman is present. A woman is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cl...
Create an eye-level scene. A rickshaw is present, facing image-right. A spoon is present. From the viewer's perspective, spoon appears closer than rickshaw. A woman is present. A woman is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear g...
Create an eye-level scene. A rickshaw is present, facing image-right. A spoon is present. The spoon is on rickshaw's own right side, from rickshaw's perspective. A woman is present. A woman is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small cl...
Create an eye-level scene. A rickshaw is present, facing image-right. A spoon is present. From the viewer's perspective, spoon appears closer than rickshaw. A woman is present. A woman is placed near the top of the image, using the viewer's image frame. All objects are shown at a similar visual size, with small clear g...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "rickshaw", "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "rickshaw", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "rickshaw", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "target_ids": [ "target", "target_2" ], "target_relations": [ "anchor_right" ], "image_level_target_id": "target_2", "for_target_label": "spoon", "image_level_target_label": "woman", "image_level_position": "image_top", "image_level_constraint": "viewer/image-frame position independent o...
{ "anchor": "rickshaw", "orientation_source_anchor": null, "target": "spoon", "targets": [ "spoon", "woman" ], "distractors": [] }
anchor_right
foreground
depth_foreground_background
anchor_right
facing_image_right
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000027
27
L1
layout_single_anchor
single_anchor
top_down
top_down
natural
object_layout
a simple object layout scene
Create a top-down map. An insect is present, facing image-right. A frog is present. The frog is behind insect, opposite insect's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. An insect is present, facing image-right. A frog is present. From the viewer's perspective, frog is positioned left of insect. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. An insect is present, facing image-right. A frog is present. The frog is behind insect, opposite insect's own facing direction. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
Create a top-down map. An insect is present, facing image-right. A frog is present. From the viewer's perspective, frog is positioned left of insect. All objects are shown at a similar visual size, with small clear gaps between them so none of the objects touch.
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "x": 128, "y": 128, "label": "insect", "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "insect", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "camera_view": "top_down", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "anchor", "label": "insect", "x": 128, "y": 128, "direction": "facing_image_right", "is_anchor": true ...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "frog" }
{ "anchor": "insect", "orientation_source_anchor": null, "target": "frog", "targets": [ "frog" ], "distractors": [] }
anchor_back
image_left
screen_left_right
anchor_back
facing_image_right
null
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
FoR-T2I-v1-000028
28
L3
layout_orientation_link
orientation_link
eye_level
eye_level
natural
object_layout
a simple object layout scene
Create an eye-level scene. A sea turtle is present, facing image-left. A rhinoceros is present. The rhinoceros faces away from the sea turtle. The rhinoceros sits on the rear side of sea turtle, using sea turtle's own orientation. A chicken is present. The chicken is behind rhinoceros, opposite rhinoceros's own facing ...
Create an eye-level scene. A sea turtle is present, facing image-left. A rhinoceros is present. The rhinoceros is facing image-right. The rhinoceros is to the right of sea turtle from the viewer's perspective. A chicken is present. The chicken is to the left of rhinoceros from the viewer's perspective. All objects are ...
Create an eye-level scene. A sea turtle is present, facing image-left. A rhinoceros is present. The rhinoceros faces away from the sea turtle. The rhinoceros sits on the rear side of sea turtle, using sea turtle's own orientation. A chicken is present. The chicken is behind rhinoceros, opposite rhinoceros's own facing ...
Create an eye-level scene. A sea turtle is present, facing image-left. A rhinoceros is present. The rhinoceros is facing image-right. The rhinoceros is to the right of sea turtle from the viewer's perspective. A chicken is present. The chicken is to the left of rhinoceros from the viewer's perspective. All objects are ...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "x": 15.36, "y": 128, "label": "sea turtle", "direction": "facing_image_left", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "sea turtle", "x": 15.36, "y": 128, "direction": "facing_image_left", "is...
{ "camera_view": "eye_level", "map": { "width": 256, "height": 256, "foreground_direction": "positive_y", "tolerance": 0.000001 }, "objects": [ { "id": "orientation_source", "label": "sea turtle", "x": 15.36, "y": 128, "direction": "facing_image_left", "is...
{ "target_ids": [ "target" ], "target_relations": [ "anchor_back" ], "for_target_label": "chicken", "orientation_source_anchor": "sea turtle", "orientation_rule": "faces_away_from_source", "orientation_source_direction": "facing_image_left", "resolved_anchor_direction": "facing_image_right", ...
{ "anchor": "rhinoceros", "orientation_source_anchor": "sea turtle", "target": "chicken", "targets": [ "chicken" ], "distractors": [], "second_anchor": "sea turtle" }
anchor_back
image_left
screen_left_right
anchor_back
facing_image_right
faces_away_from_source
null
FoR-T2I
v1_balanced_1200
FoR-T2I prompts and structured layout metadata for benchmark evaluation.
End of preview. Expand in Data Studio

FoR-T2I title art

Can Text-to-Image Models Draw from the Right Frame of Reference?

Left is not always image-left. If a person is facing the viewer, their left hand appears on the right side of the image. A text-to-image model can therefore render a plausible scene with the requested objects while still drawing the spatial relation from the wrong perspective.

FoR-T2I is the official benchmark release for Can Text-to-Image Models Draw from the Right Frame of Reference?. It evaluates whether T2I models can follow directional descriptions when the specified frame of reference differs from the camera view.

Each item pairs a frame-of-reference prompt (FoR_prompt) with a camera-view control prompt (Cam_prompt) that describes the same target layout. This paired design separates ordinary image-coordinate layout ability from the additional difficulty of using an oriented object's own left, right, front, or back. The release contains 1,200 prompt pairs across three difficulty levels, together with structured layout metadata and evaluation tools.

Across 22 recent T2I models, we observe a consistent drop from camera-view prompts to FoR prompts. We also include a training-free mitigation study based on prompt rewriting and VLM-gated visual feedback.

Results Preview

Top T2I model performance on FoR-T2I

Dataset Summary

FoR-T2I data summary

This repository provides prompts and structured layout metadata only; generated model images are not included.

Examples

FoR-T2I generated examples

Usage

from datasets import load_dataset

# After uploading this repository as a Hugging Face dataset repo:
dataset = load_dataset("ernie-research/FoR-T2I", split="test")
print(dataset[0]["FoR_prompt"])
print(dataset[0]["Cam_prompt"])

For local use without datasets:

import json
from pathlib import Path

path = Path("data/benchmark_v1_1200/FoR-T2I_1200_prompts.jsonl")
rows = [json.loads(line) for line in path.read_text().splitlines()]
print(len(rows))

Data Fields

Field Description
id Unique benchmark item identifier.
level Difficulty level, one of L1, L2, or L3.
camera_view Camera setting, either top_down or eye_level.
FoR_prompt Prompt describing the target layout through the oriented anchor object's frame of reference.
Cam_prompt Camera-view control prompt describing the same target layout with direction words from the viewer/image frame.
objects Object roles used by the prompt, including anchor and target categories.
layout_result Structured layout with coordinates, resolved orientations, and relation metadata.
layout_signature Compact evaluation-oriented representation of the same layout.
target_anchor_relation Relation of the target relative to the anchor's own orientation.
target_image_relation Equivalent target placement in the camera/image frame.
object_orientation Required anchor orientation.
orientation_rule Orientation-link rule for L3 cases, or null otherwise.
split Release split identifier.

Data Creation

Objects are drawn from existing text-to-image benchmark vocabularies and filtered for concrete, visually recognizable categories with usable orientation cues. Each item stores the resolved layout, camera view, anchor orientation, target relation, and paired prompt texts.

Evaluation

FoR-T2I can be evaluated with the scripts in the companion GitHub repository. The automatic evaluator uses open-vocabulary detection, segmentation fallback, depth cues, and VLM-based anchor-orientation verification. A prediction is counted as correct only when the required objects are detected, the target spatial relation is satisfied, and the anchor orientation is correct.

Expected generation-result rows follow this format:

{"id": "FoR-T2I-v1-000001", "prompt_kind": "FoR", "image_path": "/path/to/image.png", "status": "ok", "model": "model-name"}

prompt_kind must be either FoR or Cam; each benchmark item should have one generated image for each prompt kind.

Limitations

FoR-T2I is intended for evaluation, not training. Automatic scores depend on the selected detector, segmentation model, depth model, and VLM judge. Published results should report all generation and evaluation settings.

License

The FoR-T2I benchmark prompts and structured layout metadata are released under Creative Commons Attribution 4.0 International (CC BY 4.0). Code in the companion repository is released under the Apache License 2.0. Third-party models, checkpoints, and services remain under their own licenses and terms.

Citation

@misc{gu2026texttoimagemodelsdrawright,
  title={Can Text-to-Image Models Draw from the Right Frame of Reference?},
  author={Zheyuan Gu and Ruihang Li and Yong Huang and Yiqian Zhang and Xiangzhao Hao and Jiaxin Niu and Jiahao Hu and Zhenyu Zhang},
  year={2026},
  eprint={2608.03357},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2608.03357}
}
Downloads last month
409

Paper for ernie-research/FoR-T2I