Datasets:
image image | question string | choices list | answer string | answer_type string | auxiliary_images images list | capability_name string | question_type_name string | difficulty string | id string | sample_id string | split string | capability_id string | question_type string | template_id string | template_variant_id string | option_labels list | option_texts list | options_json string | answer_format string | metric_name string | metric_family string | scorer string | scoring_rule string | vlmeval_compatible bool | difficulty_reason string | gt_fields list | metadata_json string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Select the object categories visible in the current image from the options, with multiple selections allowed. | [
"A. truck",
"B. metal frame",
"C. white line",
"D. fence",
"E. railway"
] | ["A","B","C"] | multi_choice | visible category recognition | medium | UAV_000064_img_2306_UAV_C01_multi_choice_category_visible_multi_03 | img_2306 | test | UAV_C01 | multi_choice | UAV_C01_multi_choice_category_visible_multi_03 | UAV_C01_multi_choice_category_visible_multi_03_v5 | [
"A",
"B",
"C",
"D",
"E"
] | [
"truck",
"metal frame",
"white line",
"fence",
"railway"
] | {"A":"truck","B":"metal frame","C":"white line","D":"fence","E":"railway"} | list | Set Exact Match | Exact Match / precision / recall / F1 | uav_gt_objective_metrics.score_item | Extract all capital option letters; compare as a set and report precision/recall/F1. | true | Requires set judgment across multiple candidate categories, where the evidence chain includes visible sets and absent distractors. | [
"objects.category"
] | {"answer_evidence":{"distractors":["railway","fence"],"visible_categories":["white line","metal frame","truck"]},"answer_margin":3.0,"gt_lineage":{"answer_margin":3.0,"difficulty":"medium","difficulty_reason":"Requires set judgment across multiple candidate categories, where the evidence chain includes visible sets and... | |||
Which category of objects appears in the current frame? | [
"A. container",
"B. road",
"C. crosswalk",
"D. manhole cover"
] | D | single_choice | visible category recognition | simple | UAV_000060_img_2828_UAV_C01_single_choice_category_visible_single_07 | img_2828 | test | UAV_C01 | single_choice | UAV_C01_single_choice_category_visible_single_07 | UAV_C01_single_choice_category_visible_single_07_v2 | [
"A",
"B",
"C",
"D"
] | [
"container",
"road",
"crosswalk",
"manhole cover"
] | {"A":"container","B":"road","C":"crosswalk","D":"manhole cover"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only determine if a category belongs to the visible category set; evidence chain length is 1. | [
"objects.category"
] | {"answer_evidence":{"answer_category":"manhole cover","distractors":["container","crosswalk","road"],"visible_category_count":19},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Only determine if a category belongs to the visible category set; evidence chain length is 1.... | |||
Which interval correctly represents the visible number of low-buildings? | [
"A. 1",
"B. 10 or more",
"C. 6-10",
"D. not present",
"E. 3-5",
"F. 2"
] | E | interval_choice | Instance count and category distribution | simple | UAV_000072_img_0650_UAV_C02_interval_choice_category_count_interval_10 | img_0650 | test | UAV_C02 | interval_choice | UAV_C02_interval_choice_category_count_interval_10 | UAV_C02_interval_choice_category_count_interval_10_v1 | [
"A",
"B",
"C",
"D",
"E",
"F"
] | [
"1",
"10 or more",
"6-10",
"not present",
"3-5",
"2"
] | {"A":"1","B":"10 or more","C":"6-10","D":"not present","E":"3-5","F":"2"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single class instance count mapped to a discrete interval, evidence chain length of 1. | [
"objects.category"
] | {"answer_evidence":{"category":"low-building","count":3,"interval":"3-5"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single class instance count mapped to a discrete interval, evidence chain length of 1.","distractor_strategy":"fixed count interval bins","gt_lineage_depth":1,"involved_gt_edges":["count_to... | |||
Arrange the candidate categories from the most instances to the least. | [
"A. crop",
"B. path",
"C. trees",
"D. building"
] | ["C","D","B","A"] | ordered_list | Instance count and category distribution | hard | UAV_000036_img_4858_UAV_C02_ordered_list_category_count_order_08 | img_4858 | test | UAV_C02 | ordered_list | UAV_C02_ordered_list_category_count_order_08 | UAV_C02_ordered_list_category_count_order_08_v4 | [
"A",
"B",
"C",
"D"
] | [
"crop",
"path",
"trees",
"building"
] | {"A":"crop","B":"path","C":"trees","D":"building"} | list | Exact Order Accuracy | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons. | [
"objects.category"
] | {"answer_evidence":{"count_by_option":{"A":1,"B":7,"C":26,"D":10}},"answer_margin":3.0,"gt_lineage":{"answer_margin":3.0,"difficulty":"hard","difficulty_reason":"Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons.","distractor_strate... | |||
Which category has the largest number in the image? | [
"A. vehicle",
"B. equipment",
"C. trees",
"D. concrete"
] | D | single_choice | Instance count and category distribution | medium | UAV_000016_img_1492_UAV_C02_single_choice_category_count_max_04 | img_1492 | test | UAV_C02 | single_choice | UAV_C02_single_choice_category_count_max_04 | UAV_C02_single_choice_category_count_max_04_v1 | [
"A",
"B",
"C",
"D"
] | [
"vehicle",
"equipment",
"trees",
"concrete"
] | {"A":"vehicle","B":"equipment","C":"trees","D":"concrete"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Comparing the number of instances across multiple categories requires excluding distractors with similar counts. | [
"objects.category"
] | {"answer_evidence":{"answer_category":"concrete","count_by_category":{"concrete":32,"equipment":15,"trees":11,"vehicle":17}},"answer_margin":15.0,"gt_lineage":{"answer_margin":15.0,"difficulty":"medium","difficulty_reason":"Comparing the number of instances across multiple categories requires excluding distractors with... | |||
Please identify the image region where object A is located. | [
"A. left top",
"B. upper middle",
"C. right upper",
"D. left middle",
"E. center",
"F. right middle",
"G. left bottom",
"H. middle lower",
"I. right bottom"
] | D | grounding_choice | image plane position and area localization | simple | UAV_000055_img_3333_UAV_C03_grounding_choice_object_grid_region_02 | img_3333 | test | UAV_C03 | grounding_choice | UAV_C03_grounding_choice_object_grid_region_02 | UAV_C03_grounding_choice_object_grid_region_02_v2 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"left top",
"upper middle",
"right upper",
"left middle",
"center",
"right middle",
"left bottom",
"middle lower",
"right bottom"
] | {"A":"left top","B":"upper middle","C":"right upper","D":"left middle","E":"center","F":"right middle","G":"left bottom","H":"middle lower","I":"right bottom"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only mapping the center of a labeled object to the nine-grid area results in an evidence chain length of 1. | [
"objects.bbox_2d.centroid_xy",
"objects.bbox_2d.centroid_normalized_xy"
] | {"answer_evidence":{"centroid_normalized_xy":[0.284166,0.523367],"grid_region":"left middle","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Only mapping the center of a labeled object to the nine-grid area results in an evidence chain length of 1.","distractor_strategy":"nine-grid region o... | |||
Please write down the corresponding area option for each annotated object. | [
"A. left top",
"B. upper middle",
"C. right upper",
"D. left middle",
"E. center",
"F. right middle",
"G. left bottom",
"H. middle lower",
"I. right bottom"
] | {"A":"I","B":"F","C":"E"} | matching | image plane position and area localization | hard | UAV_000027_img_3646_UAV_C03_matching_grid_region_matching_03 | img_3646 | test | UAV_C03 | matching | UAV_C03_matching_grid_region_matching_03 | UAV_C03_matching_grid_region_matching_03_v3 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"left top",
"upper middle",
"right upper",
"left middle",
"center",
"right middle",
"left bottom",
"middle lower",
"right bottom"
] | {"A":"left top","B":"upper middle","C":"right upper","D":"left middle","E":"center","F":"right middle","G":"left bottom","H":"middle lower","I":"right bottom"} | mapping | DictExactStrMatchAggRecall | DictExactStrMatchAggRecall / matching_accuracy | uav_gt_objective_metrics.score_item | Parse a label-to-option dictionary or A:B pairs and compare each GT key. | true | Need to perform area matching for multiple annotated objects separately, with the evidence chain including multiple object-to-area mappings. | [
"objects.bbox_2d.centroid_xy",
"objects.bbox_2d.centroid_normalized_xy"
] | {"answer_evidence":{"grid_region_by_label":{"A":"right bottom","B":"right middle","C":"center"}},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Need to perform area matching for multiple annotated objects separately, with the evidence chain including multiple object-to-area mappings.","distractor_strategy":"mul... | |||
Output the annotation order from left to right based on the image horizontal position. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["A","B","D","C"] | ordered_list | image plane position and area localization | hard | UAV_000019_img_2195_UAV_C03_ordered_list_left_to_right_order_07 | img_2195 | test | UAV_C03 | ordered_list | UAV_C03_ordered_list_left_to_right_order_07 | UAV_C03_ordered_list_left_to_right_order_07_v1 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Need to do a full horizontal sorting of four annotated objects, with the evidence chain involving multiple object comparisons. | [
"objects.bbox_2d.centroid_xy"
] | {"answer_evidence":{"cx_by_label":{"A":110.295,"B":458.749,"C":1139.067,"D":662.998}},"answer_margin":204.249435,"gt_lineage":{"answer_margin":204.249435,"difficulty":"hard","difficulty_reason":"Need to do a full horizontal sorting of four annotated objects, with the evidence chain involving multiple object comparisons... | |||
Which labeled object is most centered in the observation? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | C | single_choice | image plane position and area localization | medium | UAV_000089_img_5153_UAV_C03_single_choice_nearest_center_02 | img_5153 | test | UAV_C03 | single_choice | UAV_C03_single_choice_nearest_center_02 | UAV_C03_single_choice_nearest_center_02_v5 | [
"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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the distances of the four labeled objects to the image center; evidence chain length is 2. | [
"objects.bbox_2d.centroid_xy"
] | {"answer_evidence":{"distance_to_center_by_label":{"A":864.805,"B":563.063,"C":138.314,"D":665.143}},"answer_margin":424.74847,"gt_lineage":{"answer_margin":424.74847,"difficulty":"medium","difficulty_reason":"Compare the distances of the four labeled objects to the image center; evidence chain length is 2.","distracto... | |||
Please judge the visible area scale of labeled object A in the image. | [
"A. large visible area",
"B. small visible area",
"C. medium visible area",
"D. very small visible area",
"E. larger visible area"
] | D | interval_choice | visible area and salient scale | simple | UAV_000123_img_3810_UAV_C04_interval_choice_area_interval_09 | img_3810 | test | UAV_C04 | interval_choice | UAV_C04_interval_choice_area_interval_09 | UAV_C04_interval_choice_area_interval_09_v3 | [
"A",
"B",
"C",
"D",
"E"
] | [
"large visible area",
"small visible area",
"medium visible area",
"very small visible area",
"larger visible area"
] | {"A":"large visible area","B":"small visible area","C":"medium visible area","D":"very small visible area","E":"larger visible area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single annotation instance visible area mapped to qualitative interval, answer is not bare numeric value. | [
"objects.mask.area_px",
"image_size"
] | {"answer_evidence":{"area_fraction":0.00037483,"interval":"very small visible area","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single annotation instance visible area mapped to qualitative interval, answer is not bare numeric value.","distractor_strategy":"qualitative area interval opt... | |||
Arrange these annotated objects from the largest visible area to the smallest. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["A","C","B","D"] | ordered_list | visible area and salient scale | hard | UAV_000001_img_1584_UAV_C04_ordered_list_area_order_01 | img_1584 | test | UAV_C04 | ordered_list | UAV_C04_ordered_list_area_order_01 | UAV_C04_ordered_list_area_order_01_v2 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Sort the four annotated instances by visible area, with an evidence chain length of 3. | [
"objects.mask.area_px"
] | {"answer_evidence":{"visible_area_by_label":{"A":12083,"B":6777,"C":9818,"D":3655}},"answer_margin":1.230699,"gt_lineage":{"answer_margin":1.230699,"difficulty":"hard","difficulty_reason":"Sort the four annotated instances by visible area, with an evidence chain length of 3.","distractor_strategy":"area-order ratio nea... | |||
which annotation box corresponds to the instance with the largest visible area? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | C | single_choice | visible area and salient scale | medium | UAV_000005_img_3206_UAV_C04_single_choice_largest_area_06 | img_3206 | test | UAV_C04 | single_choice | UAV_C04_single_choice_largest_area_06 | UAV_C04_single_choice_largest_area_06_v3 | [
"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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible mask areas across multiple annotated instances and use the area ratio to exclude approximate ties. | [
"objects.mask.area_px"
] | {"answer_evidence":{"visible_area_by_label":{"A":41473,"B":78809,"C":247936,"D":19166}},"answer_margin":3.146037,"gt_lineage":{"answer_margin":3.146037,"difficulty":"medium","difficulty_reason":"Compare the visible mask areas across multiple annotated instances and use the area ratio to exclude approximate ties.","dist... | |||
Select the position gear A from the near and far options. | [
"A. nearby",
"B. far away",
"C. medium distance",
"D. distant area",
"E. near and middle"
] | A | interval_choice | depth judgment | simple | UAV_000100_img_3880_UAV_C05_interval_choice_depth_interval_09 | img_3880 | test | UAV_C05 | interval_choice | UAV_C05_interval_choice_depth_interval_09 | UAV_C05_interval_choice_depth_interval_09_v4 | [
"A",
"B",
"C",
"D",
"E"
] | [
"nearby",
"far away",
"medium distance",
"distant area",
"near and middle"
] | {"A":"nearby","B":"far away","C":"medium distance","D":"distant area","E":"near and middle"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single object depth labeled and mapped to qualitative proximity intervals, answer is not a bare numerical value. | [
"objects.depth_median"
] | {"answer_evidence":{"depth":1.130924,"interval":"nearby","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single object depth labeled and mapped to qualitative proximity intervals, answer is not a bare numerical value.","distractor_strategy":"qualitative depth interval options","gt_lineage_d... | |||
Arrange the candidate objects based on the distance from the camera. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["D","A","B","C"] | ordered_list | depth judgment | hard | UAV_000052_img_2933_UAV_C05_ordered_list_depth_order_08 | img_2933 | test | UAV_C05 | ordered_list | UAV_C05_ordered_list_depth_order_08 | UAV_C05_ordered_list_depth_order_08_v1 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Fully sort the depths of the four labeled objects from nearest to farthest, with an evidence chain length of 3. | [
"objects.depth_median"
] | {"answer_evidence":{"depth_by_label":{"A":26.742298,"B":51.150703,"C":52.986839,"D":16.870714}},"answer_margin":1.836136,"gt_lineage":{"answer_margin":1.836136,"difficulty":"hard","difficulty_reason":"Fully sort the depths of the four labeled objects from nearest to farthest, with an evidence chain length of 3.","distr... | |||
Observe the annotated image. Which annotated object is closer to the camera? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | A | single_choice | depth judgment | medium | UAV_000051_img_4697_UAV_C05_single_choice_nearest_depth_03 | img_4697 | test | UAV_C05 | single_choice | UAV_C05_single_choice_nearest_depth_03 | UAV_C05_single_choice_nearest_depth_03_v5 | [
"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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the depths of multiple annotated objects and select the nearest one; approximate ties will be rejected. | [
"objects.depth_median"
] | {"answer_evidence":{"depth_by_label":{"A":8.335821,"B":16.780716,"C":18.157085,"D":17.344299}},"answer_margin":8.444895,"gt_lineage":{"answer_margin":8.444895,"difficulty":"medium","difficulty_reason":"Compare the depths of multiple annotated objects and select the nearest one; approximate ties will be rejected.","dist... | |||
Arrange the candidates in ascending order of their three-dimensional horizontal coordinates. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["A","B","C","D"] | ordered_list | Three-dimensional position in the camera coordinate system and visible three-dimensional scale | hard | UAV_000049_img_5124_UAV_C06_ordered_list_lateral_3d_order_10 | img_5124 | test | UAV_C06 | ordered_list | UAV_C06_ordered_list_lateral_3d_order_10 | UAV_C06_ordered_list_lateral_3d_order_10_v5 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Sort multiple annotated objects based on their three-dimensional lateral position, with a chain length of 3. | [
"objects.centroid_3d.xyz"
] | {"answer_evidence":{"camera_frame_x_by_label":{"A":-6.199668,"B":-4.465378,"C":0.262301,"D":2.704688}},"answer_margin":1.73429,"gt_lineage":{"answer_margin":1.73429,"difficulty":"hard","difficulty_reason":"Sort multiple annotated objects based on their three-dimensional lateral position, with a chain length of 3.","dis... | |||
Which annotated object is more to the right relative to the shooting perspective? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | C | single_choice | Three-dimensional position in the camera coordinate system and visible three-dimensional scale | medium | UAV_000006_img_4682_UAV_C06_single_choice_rightmost_3d_03 | img_4682 | test | UAV_C06 | single_choice | UAV_C06_single_choice_rightmost_3d_03 | UAV_C06_single_choice_rightmost_3d_03_v5 | [
"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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the three-dimensional lateral positions of multiple annotated objects, using margin to exclude approximate side-by-side arrangements. | [
"objects.centroid_3d.xyz"
] | {"answer_evidence":{"camera_frame_x_by_label":{"A":-1.548717,"B":-2.792789,"C":0.092465,"D":-1.945762}},"answer_margin":1.641182,"gt_lineage":{"answer_margin":1.641182,"difficulty":"medium","difficulty_reason":"Compare the three-dimensional lateral positions of multiple annotated objects, using margin to exclude approx... | |||
Which item should be selected for the shape of object A's image box? | [
"A. taller vertically",
"B. Width and height are close",
"C. wider horizontally"
] | C | interval_choice | box geometry and image boundary relationship | simple | UAV_000065_img_0414_UAV_C07_interval_choice_shape_interval_04 | img_0414 | test | UAV_C07 | interval_choice | UAV_C07_interval_choice_shape_interval_04 | UAV_C07_interval_choice_shape_interval_04_v3 | [
"A",
"B",
"C"
] | [
"taller vertically",
"Width and height are close",
"wider horizontally"
] | {"A":"taller vertically","B":"Width and height are close","C":"wider horizontally"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single aspect ratio of a bounding box mapped to a qualitative shape range, answer is not a bare numerical value. | [
"objects.bbox_2d.xyxy"
] | {"answer_evidence":{"bbox_aspect":3.210526,"object_label":"A","shape_interval":"wider horizontally"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single aspect ratio of a bounding box mapped to a qualitative shape range, answer is not a bare numerical value.","distractor_strategy":"qualitative aspect interv... | |||
Which edge is each labeled object closest to? | [
"A. left edge",
"B. right edge",
"C. top edge",
"D. bottom edge"
] | {"A":"D","B":"C","C":"D"} | matching | box geometry and image boundary relationship | hard | UAV_000002_img_4023_UAV_C07_matching_boundary_side_matching_05 | img_4023 | test | UAV_C07 | matching | UAV_C07_matching_boundary_side_matching_05 | UAV_C07_matching_boundary_side_matching_05_v3 | [
"A",
"B",
"C",
"D"
] | [
"left edge",
"right edge",
"top edge",
"bottom edge"
] | {"A":"left edge","B":"right edge","C":"top edge","D":"bottom edge"} | mapping | DictExactStrMatchAggRecall | DictExactStrMatchAggRecall / matching_accuracy | uav_gt_objective_metrics.score_item | Parse a label-to-option dictionary or A:B pairs and compare each GT key. | true | Need to match each annotated object to the nearest image edge and exclude boundaries that are nearly parallel. | [
"objects.bbox_2d.xyxy",
"image_size"
] | {"answer_evidence":{"margin_by_label":{"A":68.0,"B":119.0,"C":69.0},"nearest_boundary_by_label":{"A":"bottom edge","B":"top edge","C":"bottom edge"}},"answer_margin":68.0,"gt_lineage":{"answer_margin":68.0,"difficulty":"hard","difficulty_reason":"Need to match each annotated object to the nearest image edge and exclude... | |||
Observe the annotated image and determine which annotated object appears widest. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | A | single_choice | box geometry and image boundary relationship | medium | UAV_000008_img_0319_UAV_C07_single_choice_widest_object_06 | img_0319 | test | UAV_C07 | single_choice | UAV_C07_single_choice_widest_object_06 | UAV_C07_single_choice_widest_object_06_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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the horizontal spans of multiple bounding boxes and reject those with widths that are approximately equal. | [
"objects.bbox_2d.xyxy"
] | {"answer_evidence":{"bbox_width_by_label":{"A":880.0,"B":104.0,"C":66.0,"D":82.0}},"answer_margin":776.0,"gt_lineage":{"answer_margin":776.0,"difficulty":"medium","difficulty_reason":"Compare the horizontal spans of multiple bounding boxes and reject those with widths that are approximately equal.","distractor_strategy... | |||
Select the category of object A from the candidate categories. | [
"A. motorcycle",
"B. traffic sign",
"C. trees",
"D. tower"
] | C | marked_object_choice | Object selection and matching under category constraints | simple | UAV_000425_img_5161_UAV_C08_marked_object_choice_marked_category_choice_09 | img_5161 | test | UAV_C08 | marked_object_choice | UAV_C08_marked_object_choice_marked_category_choice_09 | UAV_C08_marked_object_choice_marked_category_choice_09_v4 | [
"A",
"B",
"C",
"D"
] | [
"motorcycle",
"traffic sign",
"trees",
"tower"
] | {"A":"motorcycle","B":"traffic sign","C":"trees","D":"tower"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only classify one annotated object, with evidence chain length of 1. | [
"objects.category",
"objects.bbox_2d.xyxy"
] | {"answer_evidence":{"answer_category":"trees","distractors":["motorcycle","traffic sign","tower"],"object_label":"A"},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Only classify one annotated object, with evidence chain length of 1.","distractor_strategy":"same-image a... | |||
Match the candidate labeled objects with object categories. | [
"A. truck",
"B. car",
"C. building"
] | {"A":"A","B":"B","C":"C"} | matching | Object selection and matching under category constraints | hard | UAV_000023_img_0829_UAV_C08_matching_category_matching_02 | img_0829 | test | UAV_C08 | matching | UAV_C08_matching_category_matching_02 | UAV_C08_matching_category_matching_02_v4 | [
"A",
"B",
"C"
] | [
"truck",
"car",
"building"
] | {"A":"truck","B":"car","C":"building"} | mapping | DictExactStrMatchAggRecall | DictExactStrMatchAggRecall / matching_accuracy | uav_gt_objective_metrics.score_item | Parse a label-to-option dictionary or A:B pairs and compare each GT key. | true | Need to match multiple annotated objects to category options, with the evidence chain including multiple object category edges. | [
"objects.category",
"objects.bbox_2d.xyxy"
] | {"answer_evidence":{"category_by_label":{"A":"truck","B":"car","C":"building"}},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"hard","difficulty_reason":"Need to match multiple annotated objects to category options, with the evidence chain including multiple object category edges.","distractor_stra... | |||
Please arrange these similar objects in order of distance. | [
"A. label object A",
"B. label object B",
"C. label object C"
] | ["A","C","B"] | ordered_list | Object selection and matching under category constraints | hard | UAV_000004_img_0007_UAV_C08_ordered_list_same_category_depth_order_06 | img_0007 | test | UAV_C08 | ordered_list | UAV_C08_ordered_list_same_category_depth_order_06 | UAV_C08_ordered_list_same_category_depth_order_06_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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Sorting multiple instances by depth under the same category condition, with the evidence chain including category filtering and depth sorting. | [
"objects.category",
"objects.depth_median"
] | {"answer_evidence":{"category":"signage","depth_by_label":{"A":15.819723,"B":32.6558,"C":16.236012}},"answer_margin":0.416288,"gt_lineage":{"answer_margin":0.416288,"difficulty":"hard","difficulty_reason":"Sorting multiple instances by depth under the same category condition, with the evidence chain including category ... | |||
Compare these annotated objects, which one is the largest? | [
"A. label object A",
"B. label object B",
"C. label object C"
] | C | single_choice | Object selection and matching under category constraints | medium | UAV_000007_img_5168_UAV_C08_single_choice_same_category_largest_04 | img_5168 | test | UAV_C08 | single_choice | UAV_C08_single_choice_same_category_largest_04 | UAV_C08_single_choice_same_category_largest_04_v3 | [
"A",
"B",
"C"
] | [
"label object A",
"label object B",
"label object C"
] | {"A":"label object A","B":"label object B","C":"label object C"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Comparing areas of multiple instances within the same category, the evidence chain involves category filtering and area comparison. | [
"objects.category",
"objects.mask.area_px"
] | {"answer_evidence":{"category":"blue tarp","visible_area_by_label":{"A":3303,"B":5713,"C":11116}},"answer_margin":1.945738,"gt_lineage":{"answer_margin":1.945738,"difficulty":"medium","difficulty_reason":"Comparing areas of multiple instances within the same category, the evidence chain involves category filtering and ... | |||
Select the overall aspect ratio type of the current image. | [
"A. vertical aspect ratio",
"B. near-square aspect ratio",
"C. horizontal aspect ratio"
] | C | interval_choice | image-level frame geometry | simple | UAV_000010_img_5622_UAV_C11_interval_choice_image_aspect_interval_06 | img_5622 | test | UAV_C11 | interval_choice | UAV_C11_interval_choice_image_aspect_interval_06 | UAV_C11_interval_choice_image_aspect_interval_06_v5 | [
"A",
"B",
"C"
] | [
"vertical aspect ratio",
"near-square aspect ratio",
"horizontal aspect ratio"
] | {"A":"vertical aspect ratio","B":"near-square aspect ratio","C":"horizontal aspect ratio"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only by mapping the aspect ratio of the entire image to a qualitative aspect ratio range, with evidence chain length of 1. | [
"image_size"
] | {"answer_evidence":{"aspect_ratio":1.333333,"image_height":3024,"image_width":4032,"interval":"horizontal aspect ratio"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Only by mapping the aspect ratio of the entire image to a qualitative aspect ratio range, with evidence chain length of 1.","distractor_strate... | |||
What is the overall direction of the current image? | [
"A. near-square aspect ratio",
"B. horizontal aspect ratio",
"C. vertical aspect ratio"
] | B | single_choice | image-level frame geometry | simple | UAV_000009_img_6127_UAV_C11_single_choice_image_orientation_choice_01 | img_6127 | test | UAV_C11 | single_choice | UAV_C11_single_choice_image_orientation_choice_01 | UAV_C11_single_choice_image_orientation_choice_01_v2 | [
"A",
"B",
"C"
] | [
"near-square aspect ratio",
"horizontal aspect ratio",
"vertical aspect ratio"
] | {"A":"near-square aspect ratio","B":"horizontal aspect ratio","C":"vertical aspect ratio"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only judge the aspect ratio of the entire image, without relying on object categories or model inference. | [
"image_size"
] | {"answer_evidence":{"aspect_ratio":1.777778,"image_height":720,"image_width":1280,"orientation":"horizontal aspect ratio"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Only judge the aspect ratio of the entire image, without relying on object categories or model inference.","distractor_strategy":"orientatio... | |||
Which candidate categories are supported by the current image? | [
"A. house",
"B. car",
"C. bicycle",
"D. building",
"E. airplane"
] | ["A","B","D"] | multi_choice | visible category recognition | medium | UAV_000113_img_5255_UAV_C01_multi_choice_category_visible_multi_07 | img_5255 | test | UAV_C01 | multi_choice | UAV_C01_multi_choice_category_visible_multi_07 | UAV_C01_multi_choice_category_visible_multi_07_v2 | [
"A",
"B",
"C",
"D",
"E"
] | [
"house",
"car",
"bicycle",
"building",
"airplane"
] | {"A":"house","B":"car","C":"bicycle","D":"building","E":"airplane"} | list | Set Exact Match | Exact Match / precision / recall / F1 | uav_gt_objective_metrics.score_item | Extract all capital option letters; compare as a set and report precision/recall/F1. | true | Requires set judgment across multiple candidate categories, where the evidence chain includes visible sets and absent distractors. | [
"objects.category"
] | {"answer_evidence":{"distractors":["bicycle","airplane"],"visible_categories":["car","house","building"]},"answer_margin":3.0,"gt_lineage":{"answer_margin":3.0,"difficulty":"medium","difficulty_reason":"Requires set judgment across multiple candidate categories, where the evidence chain includes visible sets and absent... | |||
Sort the candidate categories by the number of instances in the image. | [
"A. manufacturing plant",
"B. commercial building",
"C. trees",
"D. fence"
] | ["B","C","A","D"] | ordered_list | Instance count and category distribution | hard | UAV_000106_img_3884_UAV_C02_ordered_list_category_count_order_05 | img_3884 | test | UAV_C02 | ordered_list | UAV_C02_ordered_list_category_count_order_05 | UAV_C02_ordered_list_category_count_order_05_v2 | [
"A",
"B",
"C",
"D"
] | [
"manufacturing plant",
"commercial building",
"trees",
"fence"
] | {"A":"manufacturing plant","B":"commercial building","C":"trees","D":"fence"} | list | Exact Order Accuracy | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons. | [
"objects.category"
] | {"answer_evidence":{"count_by_option":{"A":15,"B":53,"C":33,"D":13}},"answer_margin":2.0,"gt_lineage":{"answer_margin":2.0,"difficulty":"hard","difficulty_reason":"Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons.","distractor_stra... | |||
Which category appears most densely in the figure? | [
"A. building",
"B. streetlight",
"C. bridge",
"D. tree"
] | D | single_choice | Instance count and category distribution | medium | UAV_000031_img_0878_UAV_C02_single_choice_category_count_max_08 | img_0878 | test | UAV_C02 | single_choice | UAV_C02_single_choice_category_count_max_08 | UAV_C02_single_choice_category_count_max_08_v5 | [
"A",
"B",
"C",
"D"
] | [
"building",
"streetlight",
"bridge",
"tree"
] | {"A":"building","B":"streetlight","C":"bridge","D":"tree"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Comparing the number of instances across multiple categories requires excluding distractors with similar counts. | [
"objects.category"
] | {"answer_evidence":{"answer_category":"tree","count_by_category":{"bridge":1,"building":3,"streetlight":13,"tree":68}},"answer_margin":55.0,"gt_lineage":{"answer_margin":55.0,"difficulty":"medium","difficulty_reason":"Comparing the number of instances across multiple categories requires excluding distractors with simil... | |||
According to the labeled image, which nine-grid area is A closer to? | [
"A. left top",
"B. upper middle",
"C. right upper",
"D. left middle",
"E. center",
"F. right middle",
"G. left bottom",
"H. middle lower",
"I. right bottom"
] | F | grounding_choice | image plane position and area localization | simple | UAV_000092_img_0273_UAV_C03_grounding_choice_object_grid_region_03 | img_0273 | test | UAV_C03 | grounding_choice | UAV_C03_grounding_choice_object_grid_region_03 | UAV_C03_grounding_choice_object_grid_region_03_v1 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"left top",
"upper middle",
"right upper",
"left middle",
"center",
"right middle",
"left bottom",
"middle lower",
"right bottom"
] | {"A":"left top","B":"upper middle","C":"right upper","D":"left middle","E":"center","F":"right middle","G":"left bottom","H":"middle lower","I":"right bottom"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only mapping the center of a labeled object to the nine-grid area results in an evidence chain length of 1. | [
"objects.bbox_2d.centroid_xy",
"objects.bbox_2d.centroid_normalized_xy"
] | {"answer_evidence":{"centroid_normalized_xy":[0.777217,0.34409],"grid_region":"right middle","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Only mapping the center of a labeled object to the nine-grid area results in an evidence chain length of 1.","distractor_strategy":"nine-grid region o... | |||
Which option better matches the visible area scale of A? | [
"A. small visible area",
"B. large visible area",
"C. medium visible area",
"D. very small visible area",
"E. larger visible area"
] | E | interval_choice | visible area and salient scale | simple | UAV_000427_img_3319_UAV_C04_interval_choice_area_interval_08 | img_3319 | test | UAV_C04 | interval_choice | UAV_C04_interval_choice_area_interval_08 | UAV_C04_interval_choice_area_interval_08_v2 | [
"A",
"B",
"C",
"D",
"E"
] | [
"small visible area",
"large visible area",
"medium visible area",
"very small visible area",
"larger visible area"
] | {"A":"small visible area","B":"large visible area","C":"medium visible area","D":"very small visible area","E":"larger visible area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single annotation instance visible area mapped to qualitative interval, answer is not bare numeric value. | [
"objects.mask.area_px",
"image_size"
] | {"answer_evidence":{"area_fraction":0.0147054,"interval":"larger visible area","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single annotation instance visible area mapped to qualitative interval, answer is not bare numeric value.","distractor_strategy":"qualitative area interval options"... | |||
Arrange these annotated objects from the largest visible area to the smallest. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["C","D","B","A"] | ordered_list | visible area and salient scale | hard | UAV_000022_img_3182_UAV_C04_ordered_list_area_order_03 | img_3182 | test | UAV_C04 | ordered_list | UAV_C04_ordered_list_area_order_03 | UAV_C04_ordered_list_area_order_03_v1 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Sort the four annotated instances by visible area, with an evidence chain length of 3. | [
"objects.mask.area_px"
] | {"answer_evidence":{"visible_area_by_label":{"A":2687,"B":3136,"C":13009,"D":7990}},"answer_margin":1.167101,"gt_lineage":{"answer_margin":1.167101,"difficulty":"hard","difficulty_reason":"Sort the four annotated instances by visible area, with an evidence chain length of 3.","distractor_strategy":"area-order ratio nea... | |||
which annotation box corresponds to the instance with the largest visible area? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | B | single_choice | visible area and salient scale | medium | UAV_000029_img_1770_UAV_C04_single_choice_largest_area_04 | img_1770 | test | UAV_C04 | single_choice | UAV_C04_single_choice_largest_area_04 | UAV_C04_single_choice_largest_area_04_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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible mask areas across multiple annotated instances and use the area ratio to exclude approximate ties. | [
"objects.mask.area_px"
] | {"answer_evidence":{"visible_area_by_label":{"A":3378,"B":77320,"C":2164,"D":2316}},"answer_margin":22.889284,"gt_lineage":{"answer_margin":22.889284,"difficulty":"medium","difficulty_reason":"Compare the visible mask areas across multiple annotated instances and use the area ratio to exclude approximate ties.","distra... | |||
Please judge the distance level of labeled object A. | [
"A. nearby",
"B. far away",
"C. near and middle",
"D. distant area",
"E. medium distance"
] | E | interval_choice | depth judgment | simple | UAV_000223_img_1626_UAV_C05_interval_choice_depth_interval_05 | img_1626 | test | UAV_C05 | interval_choice | UAV_C05_interval_choice_depth_interval_05 | UAV_C05_interval_choice_depth_interval_05_v5 | [
"A",
"B",
"C",
"D",
"E"
] | [
"nearby",
"far away",
"near and middle",
"distant area",
"medium distance"
] | {"A":"nearby","B":"far away","C":"near and middle","D":"distant area","E":"medium distance"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single object depth labeled and mapped to qualitative proximity intervals, answer is not a bare numerical value. | [
"objects.depth_median"
] | {"answer_evidence":{"depth":19.114361,"interval":"medium distance","object_label":"A"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Single object depth labeled and mapped to qualitative proximity intervals, answer is not a bare numerical value.","distractor_strategy":"qualitative depth interval options","gt... | |||
Arrange the labeled objects from nearest to farthest. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["C","D","B","A"] | ordered_list | depth judgment | hard | UAV_000056_img_3453_UAV_C05_ordered_list_depth_order_01 | img_3453 | test | UAV_C05 | ordered_list | UAV_C05_ordered_list_depth_order_01 | UAV_C05_ordered_list_depth_order_01_v3 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Fully sort the depths of the four labeled objects from nearest to farthest, with an evidence chain length of 3. | [
"objects.depth_median"
] | {"answer_evidence":{"depth_by_label":{"A":6.578554,"B":6.288133,"C":5.297194,"D":6.003371}},"answer_margin":0.284762,"gt_lineage":{"answer_margin":0.284762,"difficulty":"hard","difficulty_reason":"Fully sort the depths of the four labeled objects from nearest to farthest, with an evidence chain length of 3.","distracto... | |||
Output the annotation order according to the spatial horizontal layers. | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | ["A","B","C","D"] | ordered_list | Three-dimensional position in the camera coordinate system and visible three-dimensional scale | hard | UAV_000058_img_1126_UAV_C06_ordered_list_lateral_3d_order_02 | img_1126 | test | UAV_C06 | ordered_list | UAV_C06_ordered_list_lateral_3d_order_02 | UAV_C06_ordered_list_lateral_3d_order_02_v5 | [
"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 | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Sort multiple annotated objects based on their three-dimensional lateral position, with a chain length of 3. | [
"objects.centroid_3d.xyz"
] | {"answer_evidence":{"camera_frame_x_by_label":{"A":-1.053585,"B":-0.576647,"C":-0.120839,"D":1.769628}},"answer_margin":0.455808,"gt_lineage":{"answer_margin":0.455808,"difficulty":"hard","difficulty_reason":"Sort multiple annotated objects based on their three-dimensional lateral position, with a chain length of 3.","... | |||
Which of A, B, C, D is more to the right in the camera view? | [
"A. label object A",
"B. label object B",
"C. label object C",
"D. Label object D"
] | A | single_choice | Three-dimensional position in the camera coordinate system and visible three-dimensional scale | medium | UAV_000028_img_3385_UAV_C06_single_choice_rightmost_3d_05 | img_3385 | test | UAV_C06 | single_choice | UAV_C06_single_choice_rightmost_3d_05 | UAV_C06_single_choice_rightmost_3d_05_v1 | [
"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"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the three-dimensional lateral positions of multiple annotated objects, using margin to exclude approximate side-by-side arrangements. | [
"objects.centroid_3d.xyz"
] | {"answer_evidence":{"camera_frame_x_by_label":{"A":1.089378,"B":0.561297,"C":0.613338,"D":-0.973748}},"answer_margin":0.47604,"gt_lineage":{"answer_margin":0.47604,"difficulty":"medium","difficulty_reason":"Compare the three-dimensional lateral positions of multiple annotated objects, using margin to exclude approximat... | |||
Align labeled objects with the nearest image edge. | [
"A. left edge",
"B. right edge",
"C. top edge",
"D. bottom edge"
] | {"A":"C","B":"D","C":"D"} | matching | box geometry and image boundary relationship | hard | UAV_000035_img_4725_UAV_C07_matching_boundary_side_matching_05 | img_4725 | test | UAV_C07 | matching | UAV_C07_matching_boundary_side_matching_05 | UAV_C07_matching_boundary_side_matching_05_v1 | [
"A",
"B",
"C",
"D"
] | [
"left edge",
"right edge",
"top edge",
"bottom edge"
] | {"A":"left edge","B":"right edge","C":"top edge","D":"bottom edge"} | mapping | DictExactStrMatchAggRecall | DictExactStrMatchAggRecall / matching_accuracy | uav_gt_objective_metrics.score_item | Parse a label-to-option dictionary or A:B pairs and compare each GT key. | true | Need to match each annotated object to the nearest image edge and exclude boundaries that are nearly parallel. | [
"objects.bbox_2d.xyxy",
"image_size"
] | {"answer_evidence":{"margin_by_label":{"A":669.0,"B":443.0,"C":257.0},"nearest_boundary_by_label":{"A":"top edge","B":"bottom edge","C":"bottom edge"}},"answer_margin":257.0,"gt_lineage":{"answer_margin":257.0,"difficulty":"hard","difficulty_reason":"Need to match each annotated object to the nearest image edge and exc... | |||
which category describes object A? | [
"A. bicycle",
"B. railway",
"C. building",
"D. white-building"
] | D | marked_object_choice | Object selection and matching under category constraints | simple | UAV_000471_img_3717_UAV_C08_marked_object_choice_marked_category_choice_06 | img_3717 | test | UAV_C08 | marked_object_choice | UAV_C08_marked_object_choice_marked_category_choice_06 | UAV_C08_marked_object_choice_marked_category_choice_06_v3 | [
"A",
"B",
"C",
"D"
] | [
"bicycle",
"railway",
"building",
"white-building"
] | {"A":"bicycle","B":"railway","C":"building","D":"white-building"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only classify one annotated object, with evidence chain length of 1. | [
"objects.category",
"objects.bbox_2d.xyxy"
] | {"answer_evidence":{"answer_category":"white-building","distractors":["bicycle","railway","building"],"object_label":"A"},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Only classify one annotated object, with evidence chain length of 1.","distractor_strategy":"same-ima... | |||
From the width-to-height ratio, which aspect ratio does this image resemble more? | [
"A. horizontal aspect ratio",
"B. near-square aspect ratio",
"C. vertical aspect ratio"
] | A | interval_choice | image-level frame geometry | simple | UAV_000015_img_5865_UAV_C11_interval_choice_image_aspect_interval_09 | img_5865 | test | UAV_C11 | interval_choice | UAV_C11_interval_choice_image_aspect_interval_09 | UAV_C11_interval_choice_image_aspect_interval_09_v3 | [
"A",
"B",
"C"
] | [
"horizontal aspect ratio",
"near-square aspect ratio",
"vertical aspect ratio"
] | {"A":"horizontal aspect ratio","B":"near-square aspect ratio","C":"vertical aspect ratio"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | uav_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only by mapping the aspect ratio of the entire image to a qualitative aspect ratio range, with evidence chain length of 1. | [
"image_size"
] | {"answer_evidence":{"aspect_ratio":1.333333,"image_height":3024,"image_width":4032,"interval":"horizontal aspect ratio"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"Only by mapping the aspect ratio of the entire image to a qualitative aspect ratio range, with evidence chain length of 1.","distractor_strate... | |||
Which categories of objects are visible in the current image? | [
"A. overhead structure",
"B. bridge",
"C. vehicle",
"D. street light",
"E. building"
] | ["A","C","D"] | multi_choice | visible category recognition | medium | UAV_000117_img_1773_UAV_C01_multi_choice_category_visible_multi_10 | img_1773 | test | UAV_C01 | multi_choice | UAV_C01_multi_choice_category_visible_multi_10 | UAV_C01_multi_choice_category_visible_multi_10_v3 | [
"A",
"B",
"C",
"D",
"E"
] | [
"overhead structure",
"bridge",
"vehicle",
"street light",
"building"
] | {"A":"overhead structure","B":"bridge","C":"vehicle","D":"street light","E":"building"} | list | Set Exact Match | Exact Match / precision / recall / F1 | uav_gt_objective_metrics.score_item | Extract all capital option letters; compare as a set and report precision/recall/F1. | true | Requires set judgment across multiple candidate categories, where the evidence chain includes visible sets and absent distractors. | [
"objects.category"
] | {"answer_evidence":{"distractors":["building","bridge"],"visible_categories":["street light","overhead structure","vehicle"]},"answer_margin":3.0,"gt_lineage":{"answer_margin":3.0,"difficulty":"medium","difficulty_reason":"Requires set judgment across multiple candidate categories, where the evidence chain includes vis... | |||
Please sort these categories from highest to lowest quantity. | [
"A. white rectangle",
"B. scooter",
"C. pedestrian",
"D. crosswalk"
] | ["A","D","C","B"] | ordered_list | Instance count and category distribution | hard | UAV_000217_img_0285_UAV_C02_ordered_list_category_count_order_06 | img_0285 | test | UAV_C02 | ordered_list | UAV_C02_ordered_list_category_count_order_06 | UAV_C02_ordered_list_category_count_order_06_v1 | [
"A",
"B",
"C",
"D"
] | [
"white rectangle",
"scooter",
"pedestrian",
"crosswalk"
] | {"A":"white rectangle","B":"scooter","C":"pedestrian","D":"crosswalk"} | list | Exact Order Accuracy | Exact Order Accuracy / Pairwise Accuracy | uav_gt_objective_metrics.score_item | Extract capital option letters in order; require exact order and compute pairwise consistency. | true | Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons. | [
"objects.category"
] | {"answer_evidence":{"count_by_option":{"A":53,"B":11,"C":15,"D":22}},"answer_margin":4.0,"gt_lineage":{"answer_margin":4.0,"difficulty":"hard","difficulty_reason":"Need to perform full sorting across multiple categories, with evidence chain involving multiple object set counts and sorting comparisons.","distractor_stra... |
End of preview. Expand in Data Studio
BenchForge UAV
The test split contains 10,000 English visual questions over real aerial imagery. Each example links its main image and ordered auxiliary images to the full-resolution originals in this repository.
- The Hugging Face ImageFolder metadata in the root links all questions to their original images in the assets directory.
- The English JSONL records under data retain the same question IDs, choice labels, answers, image paths, and scoring definitions.
- There are 14,866 original image paths and 16,891 image references across the test split.
Stream the complete test split using Hugging Face Datasets. The browser preview can cover a subset of large ImageFolder datasets.
from datasets import load_dataset
dataset = load_dataset("jby20180901/BenchForge-uav", split="test", streaming=True)
- Downloads last month
- 82