Dataset Viewer
Auto-converted to Parquet Duplicate
image
imagewidth (px)
640
640
question
stringclasses
54 values
choices
listlengths
2
5
answer
stringclasses
45 values
answer_type
stringclasses
6 values
auxiliary_images
images listlengths
0
3
capability_name
stringclasses
8 values
question_type_name
stringclasses
1 value
difficulty
stringclasses
3 values
id
stringlengths
62
78
sample_id
stringclasses
411 values
split
stringclasses
1 value
capability_id
stringclasses
8 values
question_type
stringclasses
6 values
template_id
stringclasses
80 values
template_variant_id
stringclasses
299 values
option_labels
listlengths
2
5
option_texts
listlengths
2
5
options_json
stringclasses
188 values
answer_format
stringclasses
2 values
metric_name
stringclasses
3 values
metric_family
stringclasses
3 values
scorer
stringclasses
3 values
scoring_rule
stringclasses
3 values
vlmeval_compatible
bool
1 class
difficulty_reason
stringclasses
8 values
gt_fields
listlengths
1
3
metadata_json
stringlengths
1.27k
7.5k
Observe multiple annotated images and determine from which perspectives A appears.
[ "A. first-person view", "B. left perspective", "C. right perspective", "D. rear view" ]
["A","C"]
multi_choice
multi-view visibility consistency
medium
Town01_Opt_front_00010_CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_005011
Town01_Opt_front_00010
test
M1_cross_view_visibility
multi_choice
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_v2
[ "A", "B", "C", "D" ]
[ "first-person view", "left perspective", "right perspective", "rear view" ]
{"A":"first-person view","B":"left perspective","C":"right perspective","D":"rear view"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
same actor id matched across per-camera GT visible lists
[ "stable actor_id", "per-camera visible_actors", "per-camera bbox_2d" ]
{"answer_evidence":{"actor_id":"31394","candidate_camera_names":["front","side_left","side_right","rear"],"template_variant_id":"CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_v2","visible_camera_names":["front","side_right"]},"gt_lineage":{"answer_margin":2.0,"difficulty":"medium","difficulty_reason":"same actor id matched acr...
Which is the best observation perspective for target A?
[ "A. first-person view", "B. right perspective" ]
B
multi_view_choice
optimal observation perspective selection
medium
Town01_Opt_front_00014_CARLA_GT_M2_BEST_VIEW_SELECTION_001_007711
Town01_Opt_front_00014
test
M2_best_view_selection
multi_view_choice
CARLA_GT_M2_BEST_VIEW_SELECTION_001
CARLA_GT_M2_BEST_VIEW_SELECTION_001_v1
[ "A", "B" ]
[ "first-person view", "right perspective" ]
{"A":"first-person view","B":"right perspective"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
best view selected by largest GT visible pixel area for same actor id
[ "stable actor_id", "per-camera bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_camera":{"front":7226.0,"side_right":11573.0},"best_camera":"side_right","template_variant_id":"CARLA_GT_M2_BEST_VIEW_SELECTION_001_v1"},"gt_lineage":{"answer_margin":4347.0,"difficulty":"medium","difficulty_reason":"best view selected by largest GT visible pixel area ...
Select the road scene category for labeled object A.
[ "A. pedestrian", "B. vehicle", "C. traffic light", "D. traffic sign" ]
A
marked_object_choice
visible semantic category recognition
medium
Town01_Opt_front_00007_CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_001_008845
Town01_Opt_front_00007
test
O1_visible_semantic_category
marked_object_choice
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_001
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_001_v2
[ "A", "B", "C", "D" ]
[ "pedestrian", "vehicle", "traffic light", "traffic sign" ]
{"A":"pedestrian","B":"vehicle","C":"traffic light","D":"traffic sign"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
single marked object semantic category from simulator instance labels
[ "visible_actors[*].semantic_labels", "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"semantic_category":"pedestrian","semantic_labels":[4],"template_variant_id":"CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_001_v2","visible_area_px":436.0},"gt_lineage":{"answer_margin":436.0,"difficulty":"medium","difficulty_reason":"single marked object semantic category from simulator instance labels","...
How many people are in the image? Please choose a range.
[ "A. about 1", "B. about 4-6", "C. No visible target", "D. about 7 or more", "E. about 2-3" ]
E
interval_choice
visible object quantity range
medium
Town01_Opt_side_right_00025_CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_001_000591
Town01_Opt_side_right_00025
test
O2_visible_count_interval
interval_choice
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_001
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_001_v3
[ "A", "B", "C", "D", "E" ]
[ "about 1", "about 4-6", "No visible target", "about 7 or more", "about 2-3" ]
{"A":"about 1","B":"about 4-6","C":"No visible target","D":"about 7 or more","E":"about 2-3"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
category count mapped to interval; no bare numeric answer exposed
[ "visible_actors[*].semantic_labels", "visible_actor_count" ]
{"answer_evidence":{"category":"pedestrian","count":3,"counted_actor_ids":["31398","31396","31400"],"interval":"about 2-3","template_variant_id":"CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_001_v3"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"category count mapped to interval; no bare numeric answer exposed","distr...
Observe the annotated image and determine the horizontal order of A, B, and C in the picture.
[ "A. label object A", "B. label object B", "C. label object C" ]
["B","C","A"]
ordered_list
image plane spatial relationships
medium
Town01_Opt_side_right_00029_CARLA_GT_O3_IMAGE_PLANE_SPATIAL_001_000061
Town01_Opt_side_right_00029
test
O3_image_plane_spatial_relation
ordered_list
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_001
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_001_v2
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by bbox center x with near-tie rejection
[ "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"center_x_by_option":{"A":507.0,"B":116.0,"C":232.5},"order_axis":"image_left_to_right","template_variant_id":"CARLA_GT_O3_IMAGE_PLANE_SPATIAL_001_v2"},"gt_lineage":{"answer_margin":116.5,"difficulty":"medium","difficulty_reason":"multi-object ordering by bbox center x with near-tie rejection","dist...
In A, B, and C, which labeled objects appear larger in the image? Please rank them.
[ "A. label object A", "B. label object B", "C. label object C" ]
["C","A","B"]
ordered_list
visible area and salient scale
medium
Town05_Opt_front_00019_CARLA_GT_O4_VISIBLE_AREA_SCALE_001_007431
Town05_Opt_front_00019
test
O4_visible_area_scale
ordered_list
CARLA_GT_O4_VISIBLE_AREA_SCALE_001
CARLA_GT_O4_VISIBLE_AREA_SCALE_001_v1
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by GT visible pixel area
[ "visible_actors[*].bbox_2d.pixel_count" ]
{"answer_evidence":{"area_px_by_option":{"A":916.0,"B":282.0,"C":1238.0},"order_axis":"visible_area_desc","template_variant_id":"CARLA_GT_O4_VISIBLE_AREA_SCALE_001_v1"},"gt_lineage":{"answer_margin":322.0,"difficulty":"medium","difficulty_reason":"multi-object ordering by GT visible pixel area","distractor_strategy":"s...
Please check the categories of roads or traffic facilities that appear in the current view.
[ "A. traffic sign", "B. traffic light", "C. pedestrian walkway", "D. guardrail", "E. lane lines" ]
["E"]
multi_choice
traffic road elements recognition
medium
Town01_Opt_front_00000_CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_001_006721
Town01_Opt_front_00000
test
R1_traffic_infrastructure_semantics
multi_choice
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_001
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_001_v2
[ "A", "B", "C", "D", "E" ]
[ "traffic sign", "traffic light", "pedestrian walkway", "guardrail", "lane lines" ]
{"A":"traffic sign","B":"traffic light","C":"pedestrian walkway","D":"guardrail","E":"lane lines"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
set selection over GT semantic categories for visible traffic infrastructure
[ "semantic_histogram", "visible_actors[*].semantic_labels" ]
{"answer_evidence":{"option_categories":["traffic sign","traffic light","pedestrian walkway","guardrail","lane lines"],"present_traffic_categories":["lane lines"],"template_variant_id":"CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_001_v2"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"set selection over GT semantic ca...
Which one best matches the change in visible area of A over time?
[ "A. The visible area decreases noticeably", "B. The visible area increases noticeably", "C. basically unchanged" ]
B
temporal_choice
Temporal visibility changes
hard
Town01_Opt_front_00008_CARLA_GT_T1_TEMPORAL_VISIBILITY_001_008905
Town01_Opt_front_00008
test
T1_temporal_visibility_dynamics
temporal_choice
CARLA_GT_T1_TEMPORAL_VISIBILITY_001
CARLA_GT_T1_TEMPORAL_VISIBILITY_001_v4
[ "A", "B", "C" ]
[ "The visible area decreases noticeably", "The visible area increases noticeably", "basically unchanged" ]
{"A":"The visible area decreases noticeably","B":"The visible area increases noticeably","C":"basically unchanged"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
same actor id tracked across three time-ordered frames with visible-area margin
[ "frame_index", "stable actor_id", "bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_frame":[668.0,7384.0,4682.0],"camera_name":"front","template_variant_id":"CARLA_GT_T1_TEMPORAL_VISIBILITY_001_v4","trend":"The visible area increases noticeably"},"gt_lineage":{"answer_margin":4014.0,"difficulty":"hard","difficulty_reason":"same actor id tracked across...
Observe multiple annotated images and determine from which perspectives A appears.
[ "A. first-person view", "B. left perspective", "C. right perspective", "D. rear view" ]
["C","D"]
multi_choice
multi-view visibility consistency
medium
Town01_Opt_rear_00024_CARLA_GT_M1_CROSS_VIEW_VISIBILITY_003_006453
Town01_Opt_rear_00024
test
M1_cross_view_visibility
multi_choice
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_003
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_003_v2
[ "A", "B", "C", "D" ]
[ "first-person view", "left perspective", "right perspective", "rear view" ]
{"A":"first-person view","B":"left perspective","C":"right perspective","D":"rear view"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
same actor id matched across per-camera GT visible lists
[ "stable actor_id", "per-camera visible_actors", "per-camera bbox_2d" ]
{"answer_evidence":{"actor_id":"31396","candidate_camera_names":["front","side_left","side_right","rear"],"template_variant_id":"CARLA_GT_M1_CROSS_VIEW_VISIBILITY_003_v2","visible_camera_names":["side_right","rear"]},"gt_lineage":{"answer_margin":2.0,"difficulty":"medium","difficulty_reason":"same actor id matched acro...
Which perspective shows target A the clearest?
[ "A. rear view", "B. right perspective" ]
A
multi_view_choice
optimal observation perspective selection
medium
Town01_Opt_rear_00025_CARLA_GT_M2_BEST_VIEW_SELECTION_008_007378
Town01_Opt_rear_00025
test
M2_best_view_selection
multi_view_choice
CARLA_GT_M2_BEST_VIEW_SELECTION_008
CARLA_GT_M2_BEST_VIEW_SELECTION_008_v3
[ "A", "B" ]
[ "rear view", "right perspective" ]
{"A":"rear view","B":"right perspective"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
best view selected by largest GT visible pixel area for same actor id
[ "stable actor_id", "per-camera bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_camera":{"rear":7115.0,"side_right":1393.0},"best_camera":"rear","template_variant_id":"CARLA_GT_M2_BEST_VIEW_SELECTION_008_v3"},"gt_lineage":{"answer_margin":5722.0,"difficulty":"medium","difficulty_reason":"best view selected by largest GT visible pixel area for same...
Which of the following object categories best describes A in the image?
[ "A. traffic sign", "B. pedestrian", "C. vehicle", "D. traffic light" ]
B
marked_object_choice
visible semantic category recognition
medium
Town01_Opt_front_00020_CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_000861
Town01_Opt_front_00020
test
O1_visible_semantic_category
marked_object_choice
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_v3
[ "A", "B", "C", "D" ]
[ "traffic sign", "pedestrian", "vehicle", "traffic light" ]
{"A":"traffic sign","B":"pedestrian","C":"vehicle","D":"traffic light"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
single marked object semantic category from simulator instance labels
[ "visible_actors[*].semantic_labels", "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"semantic_category":"pedestrian","semantic_labels":[4],"template_variant_id":"CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_v3","visible_area_px":141.0},"gt_lineage":{"answer_margin":141.0,"difficulty":"medium","difficulty_reason":"single marked object semantic category from simulator instance labels","...
Which range does the number of marked vehicles belong to in the current view?
[ "A. about 7 or more", "B. about 1", "C. No visible target", "D. about 4-6", "E. about 2-3" ]
D
interval_choice
visible object quantity range
medium
Town05_Opt_rear_00008_CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_003_003929
Town05_Opt_rear_00008
test
O2_visible_count_interval
interval_choice
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_003
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_003_v2
[ "A", "B", "C", "D", "E" ]
[ "about 7 or more", "about 1", "No visible target", "about 4-6", "about 2-3" ]
{"A":"about 7 or more","B":"about 1","C":"No visible target","D":"about 4-6","E":"about 2-3"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
category count mapped to interval; no bare numeric answer exposed
[ "visible_actors[*].semantic_labels", "visible_actor_count" ]
{"answer_evidence":{"category":"vehicle","count":6,"counted_actor_ids":["43512","43514","43544","43542","43672","43695"],"interval":"about 4-6","template_variant_id":"CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_003_v2"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"category count mapped to interval; no bare numeric a...
Observe the annotated image and determine the horizontal order of A, B, and C in the picture.
[ "A. label object A", "B. label object B", "C. label object C" ]
["B","A","C"]
ordered_list
image plane spatial relationships
medium
Town10HD_Opt_front_00003_CARLA_GT_O3_IMAGE_PLANE_SPATIAL_003_005763
Town10HD_Opt_front_00003
test
O3_image_plane_spatial_relation
ordered_list
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_003
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_003_v2
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by bbox center x with near-tie rejection
[ "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"center_x_by_option":{"A":166.5,"B":59.5,"C":339.5},"order_axis":"image_left_to_right","template_variant_id":"CARLA_GT_O3_IMAGE_PLANE_SPATIAL_003_v2"},"gt_lineage":{"answer_margin":107.0,"difficulty":"medium","difficulty_reason":"multi-object ordering by bbox center x with near-tie rejection","distr...
In A, B, and C, which labeled objects appear larger in the image? Please rank them.
[ "A. label object A", "B. label object B", "C. label object C" ]
["A","B","C"]
ordered_list
visible area and salient scale
medium
Town05_Opt_top_00024_CARLA_GT_O4_VISIBLE_AREA_SCALE_002_007882
Town05_Opt_top_00024
test
O4_visible_area_scale
ordered_list
CARLA_GT_O4_VISIBLE_AREA_SCALE_002
CARLA_GT_O4_VISIBLE_AREA_SCALE_002_v1
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by GT visible pixel area
[ "visible_actors[*].bbox_2d.pixel_count" ]
{"answer_evidence":{"area_px_by_option":{"A":12478.0,"B":6233.0,"C":517.0},"order_axis":"visible_area_desc","template_variant_id":"CARLA_GT_O4_VISIBLE_AREA_SCALE_002_v1"},"gt_lineage":{"answer_margin":5716.0,"difficulty":"medium","difficulty_reason":"multi-object ordering by GT visible pixel area","distractor_strategy"...
What traffic basic elements can be seen in this image?
[ "A. traffic sign", "B. traffic light", "C. pedestrian walkway", "D. guardrail", "E. lane lines" ]
["E"]
multi_choice
traffic road elements recognition
medium
Town01_Opt_front_00001_CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_002_007362
Town01_Opt_front_00001
test
R1_traffic_infrastructure_semantics
multi_choice
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_002
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_002_v4
[ "A", "B", "C", "D", "E" ]
[ "traffic sign", "traffic light", "pedestrian walkway", "guardrail", "lane lines" ]
{"A":"traffic sign","B":"traffic light","C":"pedestrian walkway","D":"guardrail","E":"lane lines"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
set selection over GT semantic categories for visible traffic infrastructure
[ "semantic_histogram", "visible_actors[*].semantic_labels" ]
{"answer_evidence":{"option_categories":["traffic sign","traffic light","pedestrian walkway","guardrail","lane lines"],"present_traffic_categories":["lane lines"],"template_variant_id":"CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_002_v4"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"set selection over GT semantic ca...
Which one best matches the change in visible area of A over time?
[ "A. The visible area decreases noticeably", "B. basically unchanged", "C. The visible area increases noticeably" ]
C
temporal_choice
Temporal visibility changes
hard
Town01_Opt_front_00009_CARLA_GT_T1_TEMPORAL_VISIBILITY_002_001372
Town01_Opt_front_00009
test
T1_temporal_visibility_dynamics
temporal_choice
CARLA_GT_T1_TEMPORAL_VISIBILITY_002
CARLA_GT_T1_TEMPORAL_VISIBILITY_002_v4
[ "A", "B", "C" ]
[ "The visible area decreases noticeably", "basically unchanged", "The visible area increases noticeably" ]
{"A":"The visible area decreases noticeably","B":"basically unchanged","C":"The visible area increases noticeably"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
same actor id tracked across three time-ordered frames with visible-area margin
[ "frame_index", "stable actor_id", "bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_frame":[1081.0,7226.0,2466.0],"camera_name":"front","template_variant_id":"CARLA_GT_T1_TEMPORAL_VISIBILITY_002_v4","trend":"The visible area increases noticeably"},"gt_lineage":{"answer_margin":1385.0,"difficulty":"hard","difficulty_reason":"same actor id tracked acros...
In which camera views can this stable labeled target be seen?
[ "A. first-person view", "B. left perspective", "C. right perspective", "D. rear view" ]
["C","D"]
multi_choice
multi-view visibility consistency
medium
Town01_Opt_rear_00029_CARLA_GT_M1_CROSS_VIEW_VISIBILITY_008_008781
Town01_Opt_rear_00029
test
M1_cross_view_visibility
multi_choice
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_008
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_008_v4
[ "A", "B", "C", "D" ]
[ "first-person view", "left perspective", "right perspective", "rear view" ]
{"A":"first-person view","B":"left perspective","C":"right perspective","D":"rear view"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
same actor id matched across per-camera GT visible lists
[ "stable actor_id", "per-camera visible_actors", "per-camera bbox_2d" ]
{"answer_evidence":{"actor_id":"31398","candidate_camera_names":["front","side_left","side_right","rear"],"template_variant_id":"CARLA_GT_M1_CROSS_VIEW_VISIBILITY_008_v4","visible_camera_names":["side_right","rear"]},"gt_lineage":{"answer_margin":2.0,"difficulty":"medium","difficulty_reason":"same actor id matched acro...
Select the camera perspective that is most suitable for observing target A.
[ "A. left perspective", "B. first-person view" ]
A
multi_view_choice
optimal observation perspective selection
medium
Town05_Opt_front_00010_CARLA_GT_M2_BEST_VIEW_SELECTION_009_000289
Town05_Opt_front_00010
test
M2_best_view_selection
multi_view_choice
CARLA_GT_M2_BEST_VIEW_SELECTION_009
CARLA_GT_M2_BEST_VIEW_SELECTION_009_v2
[ "A", "B" ]
[ "left perspective", "first-person view" ]
{"A":"left perspective","B":"first-person view"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
best view selected by largest GT visible pixel area for same actor id
[ "stable actor_id", "per-camera bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"43484","area_px_by_camera":{"front":5466.0,"side_left":28901.0},"best_camera":"side_left","template_variant_id":"CARLA_GT_M2_BEST_VIEW_SELECTION_009_v2"},"gt_lineage":{"answer_margin":23435.0,"difficulty":"medium","difficulty_reason":"best view selected by largest GT visible pixel area f...
Select the road scene category for labeled object A.
[ "A. traffic light", "B. pedestrian", "C. traffic sign", "D. vehicle" ]
B
marked_object_choice
visible semantic category recognition
medium
Town01_Opt_rear_00034_CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_008_005738
Town01_Opt_rear_00034
test
O1_visible_semantic_category
marked_object_choice
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_008
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_008_v2
[ "A", "B", "C", "D" ]
[ "traffic light", "pedestrian", "traffic sign", "vehicle" ]
{"A":"traffic light","B":"pedestrian","C":"traffic sign","D":"vehicle"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
single marked object semantic category from simulator instance labels
[ "visible_actors[*].semantic_labels", "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"semantic_category":"pedestrian","semantic_labels":[4],"template_variant_id":"CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_008_v2","visible_area_px":218.0},"gt_lineage":{"answer_margin":218.0,"difficulty":"medium","difficulty_reason":"single marked object semantic category from simulator instance labels","...
Which range do the marked pedestrians belong to in the current view?
[ "A. about 2-3", "B. No visible target", "C. about 4-6", "D. about 1", "E. about 7 or more" ]
D
interval_choice
visible object quantity range
simple
Town01_Opt_front_00012_CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_002_004312
Town01_Opt_front_00012
test
O2_visible_count_interval
interval_choice
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_002
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_002_v3
[ "A", "B", "C", "D", "E" ]
[ "about 2-3", "No visible target", "about 4-6", "about 1", "about 7 or more" ]
{"A":"about 2-3","B":"No visible target","C":"about 4-6","D":"about 1","E":"about 7 or more"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
category count mapped to interval; no bare numeric answer exposed
[ "visible_actors[*].semantic_labels", "visible_actor_count" ]
{"answer_evidence":{"category":"pedestrian","count":1,"counted_actor_ids":["31396"],"interval":"about 1","template_variant_id":"CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_002_v3"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"category count mapped to interval; no bare numeric answer exposed","distractor_strategy":"f...
Arrange these candidate labeled targets from left to right.
[ "A. label object A", "B. label object B", "C. label object C" ]
["B","A","C"]
ordered_list
image plane spatial relationships
medium
Town10HD_Opt_side_right_00033_CARLA_GT_O3_IMAGE_PLANE_SPATIAL_004_002756
Town10HD_Opt_side_right_00033
test
O3_image_plane_spatial_relation
ordered_list
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_004
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_004_v5
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by bbox center x with near-tie rejection
[ "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"center_x_by_option":{"A":401.5,"B":281.0,"C":552.5},"order_axis":"image_left_to_right","template_variant_id":"CARLA_GT_O3_IMAGE_PLANE_SPATIAL_004_v5"},"gt_lineage":{"answer_margin":120.5,"difficulty":"medium","difficulty_reason":"multi-object ordering by bbox center x with near-tie rejection","dist...
Observe the annotated image and arrange these targets based on the size of the visible areas.
[ "A. label object A", "B. label object B", "C. label object C" ]
["A","B","C"]
ordered_list
visible area and salient scale
medium
Town10HD_Opt_side_right_00004_CARLA_GT_O4_VISIBLE_AREA_SCALE_003_001803
Town10HD_Opt_side_right_00004
test
O4_visible_area_scale
ordered_list
CARLA_GT_O4_VISIBLE_AREA_SCALE_003
CARLA_GT_O4_VISIBLE_AREA_SCALE_003_v5
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by GT visible pixel area
[ "visible_actors[*].bbox_2d.pixel_count" ]
{"answer_evidence":{"area_px_by_option":{"A":35672.0,"B":4560.0,"C":1638.0},"order_axis":"visible_area_desc","template_variant_id":"CARLA_GT_O4_VISIBLE_AREA_SCALE_003_v5"},"gt_lineage":{"answer_margin":2922.0,"difficulty":"medium","difficulty_reason":"multi-object ordering by GT visible pixel area","distractor_strategy...
Please check the categories of roads or traffic facilities that appear in the current view.
[ "A. traffic sign", "B. traffic light", "C. pedestrian walkway", "D. guardrail", "E. lane lines" ]
["E"]
multi_choice
traffic road elements recognition
medium
Town01_Opt_front_00002_CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_003_005723
Town01_Opt_front_00002
test
R1_traffic_infrastructure_semantics
multi_choice
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_003
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_003_v2
[ "A", "B", "C", "D", "E" ]
[ "traffic sign", "traffic light", "pedestrian walkway", "guardrail", "lane lines" ]
{"A":"traffic sign","B":"traffic light","C":"pedestrian walkway","D":"guardrail","E":"lane lines"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
set selection over GT semantic categories for visible traffic infrastructure
[ "semantic_histogram", "visible_actors[*].semantic_labels" ]
{"answer_evidence":{"option_categories":["traffic sign","traffic light","pedestrian walkway","guardrail","lane lines"],"present_traffic_categories":["lane lines"],"template_variant_id":"CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_003_v2"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"set selection over GT semantic ca...
Looking at the sequence of annotated images, how does the prominence of A change?
[ "A. The visible area decreases noticeably", "B. The visible area increases noticeably", "C. basically unchanged" ]
A
temporal_choice
Temporal visibility changes
hard
Town01_Opt_rear_00019_CARLA_GT_T1_TEMPORAL_VISIBILITY_003_008871
Town01_Opt_rear_00019
test
T1_temporal_visibility_dynamics
temporal_choice
CARLA_GT_T1_TEMPORAL_VISIBILITY_003
CARLA_GT_T1_TEMPORAL_VISIBILITY_003_v4
[ "A", "B", "C" ]
[ "The visible area decreases noticeably", "The visible area increases noticeably", "basically unchanged" ]
{"A":"The visible area decreases noticeably","B":"The visible area increases noticeably","C":"basically unchanged"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
same actor id tracked across three time-ordered frames with visible-area margin
[ "frame_index", "stable actor_id", "bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_frame":[4585.0,7446.0,1241.0],"camera_name":"rear","template_variant_id":"CARLA_GT_T1_TEMPORAL_VISIBILITY_003_v4","trend":"The visible area decreases noticeably"},"gt_lineage":{"answer_margin":3344.0,"difficulty":"hard","difficulty_reason":"same actor id tracked across...
Select all views that contain annotated target A.
[ "A. first-person view", "B. left perspective", "C. right perspective", "D. rear view" ]
["C","D"]
multi_choice
multi-view visibility consistency
medium
Town01_Opt_side_right_00024_CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_006511
Town01_Opt_side_right_00024
test
M1_cross_view_visibility
multi_choice
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_v3
[ "A", "B", "C", "D" ]
[ "first-person view", "left perspective", "right perspective", "rear view" ]
{"A":"first-person view","B":"left perspective","C":"right perspective","D":"rear view"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
same actor id matched across per-camera GT visible lists
[ "stable actor_id", "per-camera visible_actors", "per-camera bbox_2d" ]
{"answer_evidence":{"actor_id":"31396","candidate_camera_names":["front","side_left","side_right","rear"],"template_variant_id":"CARLA_GT_M1_CROSS_VIEW_VISIBILITY_001_v3","visible_camera_names":["side_right","rear"]},"gt_lineage":{"answer_margin":2.0,"difficulty":"medium","difficulty_reason":"same actor id matched acro...
Observe multiple annotated images, which camera perspective shows the largest visible area for A?
[ "A. left perspective", "B. right perspective", "C. first-person view" ]
C
multi_view_choice
optimal observation perspective selection
hard
Town05_Opt_front_00007_CARLA_GT_M2_BEST_VIEW_SELECTION_002_005092
Town05_Opt_front_00007
test
M2_best_view_selection
multi_view_choice
CARLA_GT_M2_BEST_VIEW_SELECTION_002
CARLA_GT_M2_BEST_VIEW_SELECTION_002_v1
[ "A", "B", "C" ]
[ "left perspective", "right perspective", "first-person view" ]
{"A":"left perspective","B":"right perspective","C":"first-person view"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
best view selected by largest GT visible pixel area for same actor id
[ "stable actor_id", "per-camera bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"43512","area_px_by_camera":{"front":49058.0,"side_left":10456.0,"side_right":11265.0},"best_camera":"front","template_variant_id":"CARLA_GT_M2_BEST_VIEW_SELECTION_002_v1"},"gt_lineage":{"answer_margin":37793.0,"difficulty":"hard","difficulty_reason":"best view selected by largest GT visi...
Observe the annotated image and determine which category of road object object A belongs to.
[ "A. traffic light", "B. pedestrian", "C. vehicle", "D. traffic sign" ]
B
marked_object_choice
visible semantic category recognition
medium
Town01_Opt_side_right_00017_CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_008551
Town01_Opt_side_right_00017
test
O1_visible_semantic_category
marked_object_choice
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_v2
[ "A", "B", "C", "D" ]
[ "traffic light", "pedestrian", "vehicle", "traffic sign" ]
{"A":"traffic light","B":"pedestrian","C":"vehicle","D":"traffic sign"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
single marked object semantic category from simulator instance labels
[ "visible_actors[*].semantic_labels", "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"semantic_category":"pedestrian","semantic_labels":[4],"template_variant_id":"CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_002_v2","visible_area_px":664.0},"gt_lineage":{"answer_margin":664.0,"difficulty":"medium","difficulty_reason":"single marked object semantic category from simulator instance labels","...
Observe the annotated image and determine in which range the visible number of pedestrians falls.
[ "A. about 4-6", "B. about 1", "C. about 2-3", "D. No visible target", "E. about 7 or more" ]
B
interval_choice
visible object quantity range
simple
Town01_Opt_front_00015_CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_004_003844
Town01_Opt_front_00015
test
O2_visible_count_interval
interval_choice
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_004
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_004_v5
[ "A", "B", "C", "D", "E" ]
[ "about 4-6", "about 1", "about 2-3", "No visible target", "about 7 or more" ]
{"A":"about 4-6","B":"about 1","C":"about 2-3","D":"No visible target","E":"about 7 or more"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
category count mapped to interval; no bare numeric answer exposed
[ "visible_actors[*].semantic_labels", "visible_actor_count" ]
{"answer_evidence":{"category":"pedestrian","count":1,"counted_actor_ids":["31396"],"interval":"about 1","template_variant_id":"CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_004_v5"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"category count mapped to interval; no bare numeric answer exposed","distractor_strategy":"f...
Please sort A, B, C, D from left to right in the image.
[ "A. label object A", "B. label object B", "C. label object C", "D. Label object D" ]
["D","B","A","C"]
ordered_list
image plane spatial relationships
hard
Town05_Opt_top_00027_CARLA_GT_O3_IMAGE_PLANE_SPATIAL_002_002212
Town05_Opt_top_00027
test
O3_image_plane_spatial_relation
ordered_list
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_002
CARLA_GT_O3_IMAGE_PLANE_SPATIAL_002_v4
[ "A", "B", "C", "D" ]
[ "label object A", "label object B", "label object C", "Label object D" ]
{"A":"label object A","B":"label object B","C":"label object C","D":"Label object D"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by bbox center x with near-tie rejection
[ "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"center_x_by_option":{"A":319.5,"B":229.0,"C":479.5,"D":91.5},"order_axis":"image_left_to_right","template_variant_id":"CARLA_GT_O3_IMAGE_PLANE_SPATIAL_002_v4"},"gt_lineage":{"answer_margin":90.5,"difficulty":"hard","difficulty_reason":"multi-object ordering by bbox center x with near-tie rejection"...
Please sort these annotated targets based on the size of their visible range.
[ "A. label object A", "B. label object B", "C. label object C" ]
["B","C","A"]
ordered_list
visible area and salient scale
medium
Town10HD_Opt_top_00004_CARLA_GT_O4_VISIBLE_AREA_SCALE_004_004764
Town10HD_Opt_top_00004
test
O4_visible_area_scale
ordered_list
CARLA_GT_O4_VISIBLE_AREA_SCALE_004
CARLA_GT_O4_VISIBLE_AREA_SCALE_004_v2
[ "A", "B", "C" ]
[ "label object A", "label object B", "label object C" ]
{"A":"label object A","B":"label object B","C":"label object C"}
list
Exact Order Accuracy + Pairwise Accuracy
ranking
carla_gt_objective_metrics.ordered_scores
Full credit for exact option permutation; diagnostic pairwise consistency for all ordered pairs.
true
multi-object ordering by GT visible pixel area
[ "visible_actors[*].bbox_2d.pixel_count" ]
{"answer_evidence":{"area_px_by_option":{"A":2462.0,"B":6401.0,"C":5404.0},"order_axis":"visible_area_desc","template_variant_id":"CARLA_GT_O4_VISIBLE_AREA_SCALE_004_v2"},"gt_lineage":{"answer_margin":997.0,"difficulty":"medium","difficulty_reason":"multi-object ordering by GT visible pixel area","distractor_strategy":...
What traffic basic elements can be seen in this image?
[ "A. traffic sign", "B. traffic light", "C. pedestrian walkway", "D. guardrail", "E. lane lines" ]
["E"]
multi_choice
traffic road elements recognition
medium
Town01_Opt_front_00003_CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_004_004164
Town01_Opt_front_00003
test
R1_traffic_infrastructure_semantics
multi_choice
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_004
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_004_v4
[ "A", "B", "C", "D", "E" ]
[ "traffic sign", "traffic light", "pedestrian walkway", "guardrail", "lane lines" ]
{"A":"traffic sign","B":"traffic light","C":"pedestrian walkway","D":"guardrail","E":"lane lines"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
set selection over GT semantic categories for visible traffic infrastructure
[ "semantic_histogram", "visible_actors[*].semantic_labels" ]
{"answer_evidence":{"option_categories":["traffic sign","traffic light","pedestrian walkway","guardrail","lane lines"],"present_traffic_categories":["lane lines"],"template_variant_id":"CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_004_v4"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"set selection over GT semantic ca...
Looking at the sequence of annotated images, how does the prominence of A change?
[ "A. The visible area increases noticeably", "B. The visible area decreases noticeably", "C. basically unchanged" ]
B
temporal_choice
Temporal visibility changes
hard
Town01_Opt_rear_00020_CARLA_GT_T1_TEMPORAL_VISIBILITY_004_003404
Town01_Opt_rear_00020
test
T1_temporal_visibility_dynamics
temporal_choice
CARLA_GT_T1_TEMPORAL_VISIBILITY_004
CARLA_GT_T1_TEMPORAL_VISIBILITY_004_v4
[ "A", "B", "C" ]
[ "The visible area increases noticeably", "The visible area decreases noticeably", "basically unchanged" ]
{"A":"The visible area increases noticeably","B":"The visible area decreases noticeably","C":"basically unchanged"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
same actor id tracked across three time-ordered frames with visible-area margin
[ "frame_index", "stable actor_id", "bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"31396","area_px_by_frame":[5950.0,7115.0,793.0],"camera_name":"rear","template_variant_id":"CARLA_GT_T1_TEMPORAL_VISIBILITY_004_v4","trend":"The visible area decreases noticeably"},"gt_lineage":{"answer_margin":5157.0,"difficulty":"hard","difficulty_reason":"same actor id tracked across ...
A which perspectives are visible in the current multi-view image?
[ "A. first-person view", "B. left perspective", "C. right perspective", "D. rear view" ]
["A","C","D"]
multi_choice
multi-view visibility consistency
hard
Town05_Opt_front_00008_CARLA_GT_M1_CROSS_VIEW_VISIBILITY_002_001227
Town05_Opt_front_00008
test
M1_cross_view_visibility
multi_choice
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_002
CARLA_GT_M1_CROSS_VIEW_VISIBILITY_002_v3
[ "A", "B", "C", "D" ]
[ "first-person view", "left perspective", "right perspective", "rear view" ]
{"A":"first-person view","B":"left perspective","C":"right perspective","D":"rear view"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
same actor id matched across per-camera GT visible lists
[ "stable actor_id", "per-camera visible_actors", "per-camera bbox_2d" ]
{"answer_evidence":{"actor_id":"43512","candidate_camera_names":["front","side_left","side_right","rear"],"template_variant_id":"CARLA_GT_M1_CROSS_VIEW_VISIBILITY_002_v3","visible_camera_names":["front","side_right","rear"]},"gt_lineage":{"answer_margin":3.0,"difficulty":"hard","difficulty_reason":"same actor id matche...
In which view does A occupy the largest area of the image?
[ "A. left perspective", "B. rear view", "C. first-person view" ]
A
multi_view_choice
optimal observation perspective selection
hard
Town05_Opt_front_00009_CARLA_GT_M2_BEST_VIEW_SELECTION_003_005203
Town05_Opt_front_00009
test
M2_best_view_selection
multi_view_choice
CARLA_GT_M2_BEST_VIEW_SELECTION_003
CARLA_GT_M2_BEST_VIEW_SELECTION_003_v4
[ "A", "B", "C" ]
[ "left perspective", "rear view", "first-person view" ]
{"A":"left perspective","B":"rear view","C":"first-person view"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
best view selected by largest GT visible pixel area for same actor id
[ "stable actor_id", "per-camera bbox_2d.pixel_count" ]
{"answer_evidence":{"actor_id":"37411","area_px_by_camera":{"front":34461.0,"rear":8892.0,"side_left":42748.0},"best_camera":"side_left","template_variant_id":"CARLA_GT_M2_BEST_VIEW_SELECTION_003_v4"},"gt_lineage":{"answer_margin":8287.0,"difficulty":"hard","difficulty_reason":"best view selected by largest GT visible ...
Which of the following object categories best describes A in the image?
[ "A. vehicle", "B. pedestrian", "C. traffic light", "D. traffic sign" ]
B
marked_object_choice
visible semantic category recognition
medium
Town01_Opt_side_right_00026_CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_004_001414
Town01_Opt_side_right_00026
test
O1_visible_semantic_category
marked_object_choice
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_004
CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_004_v3
[ "A", "B", "C", "D" ]
[ "vehicle", "pedestrian", "traffic light", "traffic sign" ]
{"A":"vehicle","B":"pedestrian","C":"traffic light","D":"traffic sign"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
single marked object semantic category from simulator instance labels
[ "visible_actors[*].semantic_labels", "visible_actors[*].bbox_2d" ]
{"answer_evidence":{"semantic_category":"pedestrian","semantic_labels":[4],"template_variant_id":"CARLA_GT_O1_VISIBLE_SEMANTIC_CATEGORY_004_v3","visible_area_px":606.0},"gt_lineage":{"answer_margin":606.0,"difficulty":"medium","difficulty_reason":"single marked object semantic category from simulator instance labels","...
Observe the annotated image and determine in which range the visible number of pedestrians falls.
[ "A. No visible target", "B. about 2-3", "C. about 1", "D. about 7 or more", "E. about 4-6" ]
C
interval_choice
visible object quantity range
simple
Town01_Opt_front_00017_CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_005_001175
Town01_Opt_front_00017
test
O2_visible_count_interval
interval_choice
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_005
CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_005_v5
[ "A", "B", "C", "D", "E" ]
[ "No visible target", "about 2-3", "about 1", "about 7 or more", "about 4-6" ]
{"A":"No visible target","B":"about 2-3","C":"about 1","D":"about 7 or more","E":"about 4-6"}
scalar
Accuracy
Accuracy / exact_matching
carla_gt_objective_metrics.exact_match
Extract the first option letter and compare with the GT option key.
true
category count mapped to interval; no bare numeric answer exposed
[ "visible_actors[*].semantic_labels", "visible_actor_count" ]
{"answer_evidence":{"category":"pedestrian","count":1,"counted_actor_ids":["31396"],"interval":"about 1","template_variant_id":"CARLA_GT_O2_VISIBLE_COUNT_INTERVAL_005_v5"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"category count mapped to interval; no bare numeric answer exposed","distractor_strategy":"f...
From the annotated image, which traffic-related elements are visible?
[ "A. traffic sign", "B. traffic light", "C. pedestrian walkway", "D. guardrail", "E. lane lines" ]
["E"]
multi_choice
traffic road elements recognition
medium
Town01_Opt_front_00004_CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_005_003528
Town01_Opt_front_00004
test
R1_traffic_infrastructure_semantics
multi_choice
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_005
CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_005_v1
[ "A", "B", "C", "D", "E" ]
[ "traffic sign", "traffic light", "pedestrian walkway", "guardrail", "lane lines" ]
{"A":"traffic sign","B":"traffic light","C":"pedestrian walkway","D":"guardrail","E":"lane lines"}
list
Set Exact Match + F1
Exact Match / precision / recall / F1
carla_gt_objective_metrics.set_scores
Compare predicted option set with GT option set; report exact match, precision, recall, F1.
true
set selection over GT semantic categories for visible traffic infrastructure
[ "semantic_histogram", "visible_actors[*].semantic_labels" ]
{"answer_evidence":{"option_categories":["traffic sign","traffic light","pedestrian walkway","guardrail","lane lines"],"present_traffic_categories":["lane lines"],"template_variant_id":"CARLA_GT_R1_TRAFFIC_INFRASTRUCTURE_005_v1"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"set selection over GT semantic ca...
End of preview. Expand in Data Studio

BenchForge CARLA

Road-scene understanding and reasoning across CARLA driving views.

The test split contains 10,000 English questions. The English edition preserves the original question IDs, option labels, answers, and scoring definitions. The main image and all referenced auxiliary images are embedded in the Parquet files as their original image bytes.

  • image: Main image supplied to the question.
  • auxiliary_images: Ordered list of additional images supplied with the question. An empty list means the task has no additional images.
  • question, choices, and answer: English question and answer choices, with the original option labels in the scoring answer.
  • data/test-00000-of-00001.jsonl: English text records with their original relative image paths for users of the existing benchmark layout.

Load the English test split using Hugging Face Datasets:

from datasets import load_dataset

dataset = load_dataset("jby20180901/BenchForge-carla", split="test")

Use the Parquet edition to access images directly. For the JSONL edition, use the repository’s assets/ directory or extract assets.tar at the repository root if the dataset ships with an image archive.

Downloads last month
76