Datasets:
image imagewidth (px) 640 918 | question stringclasses 65
values | choices listlengths 3 9 | answer stringclasses 39
values | answer_type stringclasses 2
values | auxiliary_images images listlengths 1 4 | capability_name stringclasses 7
values | question_type_name stringclasses 4
values | difficulty stringclasses 3
values | id stringlengths 41 60 | sample_id stringclasses 870
values | split stringclasses 1
value | capability_id stringclasses 7
values | question_type stringclasses 4
values | template_id stringclasses 13
values | template_variant_id stringclasses 65
values | option_labels listlengths 3 9 | option_texts listlengths 3 9 | options_json stringclasses 598
values | answer_format stringclasses 2
values | metric_name stringclasses 2
values | metric_family stringclasses 2
values | scorer stringclasses 2
values | scoring_rule stringclasses 2
values | vlmeval_compatible bool 1
class | difficulty_reason stringclasses 13
values | gt_fields listlengths 2 2 | metadata_json stringlengths 1.94k 4.64k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Observe the annotated image and determine which area among A, B, C, and D is overall closer to the camera. | [
"A. left upper area",
"B. right lower area",
"C. right upper area",
"D. left bottom area"
] | A | single_choice | single-frame regional depth localization | Location Selection Question | simple | apartment_1_HAB_C01_Q01_NEAREST_QUADRANT_0007 | apartment_1:00024 | test | HAB_C01 | grounding_choice | HAB_C01_Q01_NEAREST_QUADRANT | HAB_C01_Q01_NEAREST_QUADRANT_v01 | [
"A",
"B",
"C",
"D"
] | [
"left upper area",
"right lower area",
"right upper area",
"left bottom area"
] | {"A":"left upper area","B":"right lower area","C":"right upper area","D":"left bottom area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single frame, single field depth statistics, four-out-of-one region | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":2.41566,"B":3.81994,"C":3.72885,"D":2.8687}},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single frame, single field depth statistics, four-out-of-one region","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmin"],"involved_gt_nodes":["frame","depth... | ||
Which area is closer and which is farther? Please answer from near to far. | [
"A. right region",
"B. central area",
"C. left area"
] | ["A","C","B"] | ordered_list | single-frame regional depth localization | Sorting Question | medium | apartment_1_HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_0005 | apartment_1:00155 | test | HAB_C01 | ordered_list | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_v05 | [
"A",
"B",
"C"
] | [
"right region",
"central area",
"left area"
] | {"A":"right region","B":"central area","C":"left area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Single frame with three object regions sorted, evidence chain includes three depth statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_near_to_far":["A","C","B"]},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Single frame with three object regions sorted, evidence chain includes three depth statistics.","gt_lineage_depth":5,"involved_gt_edges":["median","sort_ascending"],"involved_gt_nodes":["frame","depth_pixel... | ||
Observe the annotated image and select the closest prominent segment. | [
"A. left upper area",
"B. left bottom area",
"C. right lower area",
"D. right upper area"
] | C | single_choice | near-field spatial coverage | Location Selection Question | medium | apartment_1_HAB_C02_Q01_NEAR_SPACE_QUADRANT_0000 | apartment_1:00162 | test | HAB_C02 | grounding_choice | HAB_C02_Q01_NEAR_SPACE_QUADRANT | HAB_C02_Q01_NEAR_SPACE_QUADRANT_v03 | [
"A",
"B",
"C",
"D"
] | [
"left upper area",
"left bottom area",
"right lower area",
"right upper area"
] | {"A":"left upper area","B":"left bottom area","C":"right lower area","D":"right upper area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare the region sets by measuring the proportion of each covered by nearby content. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.07453,"B":0.29092,"C":0.65085,"D":0.59703}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Compare the region sets by measuring the proportion of each covered by nearby content.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_no... | ||
Compare the proportion of nearby content in the four regions, and rank them from highest to lowest. | [
"A. left bottom area",
"B. right upper area",
"C. left upper area",
"D. right lower area"
] | ["D","A","C","B"] | ordered_list | near-field spatial coverage | Sorting Question | hard | apartment_1_HAB_C02_Q02_NEAR_SPACE_ORDER_0001 | apartment_1:00308 | test | HAB_C02 | ordered_list | HAB_C02_Q02_NEAR_SPACE_ORDER | HAB_C02_Q02_NEAR_SPACE_ORDER_v03 | [
"A",
"B",
"C",
"D"
] | [
"left bottom area",
"right upper area",
"left upper area",
"right lower area"
] | {"A":"left bottom area","B":"right upper area","C":"left upper area","D":"right lower area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Rank four region sets in a single frame using four measurements of nearby coverage. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_high_to_low_near_fraction":["D","A","C","B"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Rank four region sets in a single frame using four measurements of nearby coverage.","gt_lineage_depth":5,"involved_gt_edges":["near_threshold_fraction","sort_descending"],"involved_gt_nodes... | ||
Please select the area with the richest layering in the image. | [
"A. central area",
"B. right region",
"C. left area"
] | A | single_choice | depth undulation and transparency | Multiple Choice Question | medium | apartment_1_HAB_C03_Q01_DEPTH_VARIATION_REGION_0013 | apartment_1:00317 | test | HAB_C03 | single_choice | HAB_C03_Q01_DEPTH_VARIATION_REGION | HAB_C03_Q01_DEPTH_VARIATION_REGION_v05 | [
"A",
"B",
"C"
] | [
"central area",
"right region",
"left area"
] | {"A":"central area","B":"right region","C":"left area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Comparing regional geometric variations in a single frame requires quantile statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.36728,"B":0.29068,"C":0.15656}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Comparing regional geometric variations in a single frame requires quantile statistics.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["fram... | ||
Compare T1→T2, T2→T3, and T3→T4, which step is the longest in the top view? | [
"A. T3→T4",
"B. T1→T2",
"C. T2→T3"
] | C | single_choice | camera plane trajectory | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | hard | apartment_1_HAB_C05_Q01_LARGEST_STEP_0000 | apartment_1:00081 | test | HAB_C05 | multi_view_choice | HAB_C05_Q01_LARGEST_STEP | HAB_C05_Q01_LARGEST_STEP_v02 | [
"A",
"B",
"C"
] | [
"T3→T4",
"T1→T2",
"T2→T3"
] | {"A":"T3→T4","B":"T1→T2","C":"T2→T3"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare trajectory diagrams in three segments, with evidence spanning position nodes. | [
"records.sensor_states.depth_sensor.position",
"records.frame_index"
] | {"answer_evidence":{"largest_step":"T2→T3"},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Compare trajectory diagrams in three segments, with evidence spanning position nodes.","gt_lineage_depth":4,"involved_gt_edges":["planar_distance","argmax"],"involved_gt_nodes":["ordered_frames","sensor_positions"]},"invo... | ||
Which direction is the overall displacement from T1 to T4 closer to? | [
"A. north-facing",
"B. East-facing",
"C. West",
"D. south-facing"
] | A | single_choice | camera plane trajectory | Multiple Choice Question | medium | apartment_1_HAB_C05_Q02_NET_DIRECTION_0005 | apartment_1:00209 | test | HAB_C05 | single_choice | HAB_C05_Q02_NET_DIRECTION | HAB_C05_Q02_NET_DIRECTION_v03 | [
"A",
"B",
"C",
"D"
] | [
"north-facing",
"East-facing",
"West",
"south-facing"
] | {"A":"north-facing","B":"East-facing","C":"West","D":"south-facing"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | summarize the four-point trajectory into a net direction. | [
"records.sensor_states.depth_sensor.position",
"records.frame_index"
] | {"answer_evidence":{"net_direction":"north-facing"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"summarize the four-point trajectory into a net direction.","gt_lineage_depth":4,"involved_gt_edges":["net_displacement","cardinal_bin"],"involved_gt_nodes":["ordered_frames","sensor_positions"]},"involved_object... | ||
Only look at the top-down view arrows and choose the most appropriate turning description. | [
"A. top view, rotate clockwise",
"B. near reverse rotation",
"C. top view rotating counterclockwise",
"D. basically maintain original direction"
] | B | single_choice | camera orientation change | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | medium | apartment_1_HAB_C06_Q01_HEADING_CHANGE_0003 | apartment_1:00227 | test | HAB_C06 | multi_view_choice | HAB_C06_Q01_HEADING_CHANGE | HAB_C06_Q01_HEADING_CHANGE_v04 | [
"A",
"B",
"C",
"D"
] | [
"top view, rotate clockwise",
"near reverse rotation",
"top view rotating counterclockwise",
"basically maintain original direction"
] | {"A":"top view, rotate clockwise","B":"near reverse rotation","C":"top view rotating counterclockwise","D":"basically maintain original direction"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | two time points with direction classes that are different. | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"turn_class":"near reverse rotation"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"two time points with direction classes that are different.","gt_lineage_depth":5,"involved_gt_edges":["forward_vector","signed_angle_delta","turn_class"],"involved_gt_nodes":["ordered_frames","sensor_rotat... | ||
Which direction is the camera at point T1 mainly facing? | [
"A. East-facing",
"B. south-facing",
"C. West",
"D. north-facing"
] | C | single_choice | camera orientation change | Multiple Choice Question | simple | apartment_1_HAB_C06_Q02_CARDINAL_HEADING_0000 | apartment_1:00271 | test | HAB_C06 | single_choice | HAB_C06_Q02_CARDINAL_HEADING | HAB_C06_Q02_CARDINAL_HEADING_v05 | [
"A",
"B",
"C",
"D"
] | [
"East-facing",
"south-facing",
"West",
"north-facing"
] | {"A":"East-facing","B":"south-facing","C":"West","D":"north-facing"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single time step pose field to direction category | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"cardinal_heading":"West"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single time step pose field to direction category","gt_lineage_depth":4,"involved_gt_edges":["forward_vector","cardinal_bin"],"involved_gt_nodes":["frame","sensor_rotation"]},"involved_objects":[],"provenance":{"gt_f... | ||
Among nine labeled grids, which grid is closest to the lens overall? | [
"A. left top cell",
"B. Middle-right cell",
"C. middle lower slot",
"D. Upper-right cell",
"E. middle upper slot",
"F. left bottom cell",
"G. Center cell",
"H. Middle-left cell",
"I. Lower-right cell"
] | I | single_choice | image plane region localization | Location Selection Question | medium | apartment_1_HAB_C07_Q01_GRID_NEAREST_CELL_0001 | apartment_1:00311 | test | HAB_C07 | grounding_choice | HAB_C07_Q01_GRID_NEAREST_CELL | HAB_C07_Q01_GRID_NEAREST_CELL_v01 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"left top cell",
"Middle-right cell",
"middle lower slot",
"Upper-right cell",
"middle upper slot",
"left bottom cell",
"Center cell",
"Middle-left cell",
"Lower-right cell"
] | {"A":"left top cell","B":"Middle-right cell","C":"middle lower slot","D":"Upper-right cell","E":"middle upper slot","F":"left bottom cell","G":"Center cell","H":"Middle-left cell","I":"Lower-right cell"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statistics. | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":2.62516,"B":1.96338,"C":2.53498,"D":1.97238,"E":2.37064,"F":2.92454,"G":2.561,"H":3.11025,"I":1.82788}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statistic... | ||
Which labeled area in the image has the most obvious distant view? | [
"A. right region",
"B. central area",
"C. left area"
] | B | single_choice | image plane region localization | Multiple Choice Question | medium | apartment_1_HAB_C07_Q02_FAR_SPACE_COLUMN_0003 | apartment_1:00159 | test | HAB_C07 | single_choice | HAB_C07_Q02_FAR_SPACE_COLUMN | HAB_C07_Q02_FAR_SPACE_COLUMN_v05 | [
"A",
"B",
"C"
] | [
"right region",
"central area",
"left area"
] | {"A":"right region","B":"central area","C":"left area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame region localization and far-field proportion statistics | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":0.0,"B":0.34817,"C":0.19107}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame region localization and far-field proportion statistics","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["frame","depth_pixels","regi... | ||
View the top-down trajectory on the right side. Please sort T2, T3, T4 from closest to farthest from T1. | [
"A. Time T2",
"B. Time T3",
"C. Time T4"
] | ["A","B","C"] | ordered_list | order of collection and multi-view consistency | Sorting Question | hard | apartment_1_HAB_C08_Q03_DISTANCE_FROM_START_ORDER_0000 | apartment_1:00323 | test | HAB_C08 | ordered_list | HAB_C08_Q03_DISTANCE_FROM_START_ORDER | HAB_C08_Q03_DISTANCE_FROM_START_ORDER_v01 | [
"A",
"B",
"C"
] | [
"Time T2",
"Time T3",
"Time T4"
] | {"A":"Time T2","B":"Time T3","C":"Time T4"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes. | [
"records.frame_index",
"records.sensor_states.depth_sensor.position"
] | {"answer_evidence":{"ordered_by_distance_from_T1":["A","B","C"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes.","gt_lineage_depth":4,"involved_gt_edges":["distance_to_reference","sort_ascending"],"involved... | ||
Which area's visible surface is closest to the camera? | [
"A. right lower area",
"B. left upper area",
"C. right upper area",
"D. left bottom area"
] | B | single_choice | single-frame regional depth localization | Location Selection Question | simple | apartment_1_HAB_C01_Q01_NEAREST_QUADRANT_0011 | apartment_1:00258 | test | HAB_C01 | grounding_choice | HAB_C01_Q01_NEAREST_QUADRANT | HAB_C01_Q01_NEAREST_QUADRANT_v05 | [
"A",
"B",
"C",
"D"
] | [
"right lower area",
"left upper area",
"right upper area",
"left bottom area"
] | {"A":"right lower area","B":"left upper area","C":"right upper area","D":"left bottom area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single frame, single field depth statistics, four-out-of-one region | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":2.95537,"B":2.44862,"C":3.0779,"D":2.66972}},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single frame, single field depth statistics, four-out-of-one region","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmin"],"involved_gt_nodes":["frame","depth... | ||
Please sort A, B, and C areas from closest to farthest from the camera. | [
"A. left area",
"B. central area",
"C. right region"
] | ["B","C","A"] | ordered_list | single-frame regional depth localization | Sorting Question | medium | apartment_1_HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_0006 | apartment_1:00195 | test | HAB_C01 | ordered_list | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_v01 | [
"A",
"B",
"C"
] | [
"left area",
"central area",
"right region"
] | {"A":"left area","B":"central area","C":"right region"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Single frame with three object regions sorted, evidence chain includes three depth statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_near_to_far":["B","C","A"]},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Single frame with three object regions sorted, evidence chain includes three depth statistics.","gt_lineage_depth":5,"involved_gt_edges":["median","sort_ascending"],"involved_gt_nodes":["frame","depth_pixel... | ||
Which region in the image has the most foreground? | [
"A. left bottom area",
"B. right upper area",
"C. left upper area",
"D. right lower area"
] | A | single_choice | near-field spatial coverage | Location Selection Question | medium | apartment_1_HAB_C02_Q01_NEAR_SPACE_QUADRANT_0017 | apartment_1:00084 | test | HAB_C02 | grounding_choice | HAB_C02_Q01_NEAR_SPACE_QUADRANT | HAB_C02_Q01_NEAR_SPACE_QUADRANT_v05 | [
"A",
"B",
"C",
"D"
] | [
"left bottom area",
"right upper area",
"left upper area",
"right lower area"
] | {"A":"left bottom area","B":"right upper area","C":"left upper area","D":"right lower area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare the region sets by measuring the proportion of each covered by nearby content. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.67614,"B":0.06654,"C":0.63917,"D":0.54071}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Compare the region sets by measuring the proportion of each covered by nearby content.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_no... | ||
Compare the proportion of nearby content in the four regions, and rank them from highest to lowest. | [
"A. right lower area",
"B. left upper area",
"C. right upper area",
"D. left bottom area"
] | ["A","C","B","D"] | ordered_list | near-field spatial coverage | Sorting Question | hard | apartment_1_HAB_C02_Q02_NEAR_SPACE_ORDER_0006 | apartment_1:00315 | test | HAB_C02 | ordered_list | HAB_C02_Q02_NEAR_SPACE_ORDER | HAB_C02_Q02_NEAR_SPACE_ORDER_v03 | [
"A",
"B",
"C",
"D"
] | [
"right lower area",
"left upper area",
"right upper area",
"left bottom area"
] | {"A":"right lower area","B":"left upper area","C":"right upper area","D":"left bottom area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Rank four region sets in a single frame using four measurements of nearby coverage. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_high_to_low_near_fraction":["A","C","B","D"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Rank four region sets in a single frame using four measurements of nearby coverage.","gt_lineage_depth":5,"involved_gt_edges":["near_threshold_fraction","sort_descending"],"involved_gt_nodes... | ||
which labeled area can see both closer and more distant content? | [
"A. central area",
"B. right region",
"C. left area"
] | A | single_choice | depth undulation and transparency | Multiple Choice Question | medium | apartment_1_HAB_C03_Q01_DEPTH_VARIATION_REGION_0025 | apartment_1:00214 | test | HAB_C03 | single_choice | HAB_C03_Q01_DEPTH_VARIATION_REGION | HAB_C03_Q01_DEPTH_VARIATION_REGION_v02 | [
"A",
"B",
"C"
] | [
"central area",
"right region",
"left area"
] | {"A":"central area","B":"right region","C":"left area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Comparing regional geometric variations in a single frame requires quantile statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.93751,"B":0.31113,"C":0.85573}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Comparing regional geometric variations in a single frame requires quantile statistics.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["fram... | ||
Select the longest adjacent movement segment in the top-down trajectory. | [
"A. T2→T3",
"B. T3→T4",
"C. T1→T2"
] | B | single_choice | camera plane trajectory | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | hard | apartment_1_HAB_C05_Q01_LARGEST_STEP_0008 | apartment_1:00251 | test | HAB_C05 | multi_view_choice | HAB_C05_Q01_LARGEST_STEP | HAB_C05_Q01_LARGEST_STEP_v05 | [
"A",
"B",
"C"
] | [
"T2→T3",
"T3→T4",
"T1→T2"
] | {"A":"T2→T3","B":"T3→T4","C":"T1→T2"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare trajectory diagrams in three segments, with evidence spanning position nodes. | [
"records.sensor_states.depth_sensor.position",
"records.frame_index"
] | {"answer_evidence":{"largest_step":"T3→T4"},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Compare trajectory diagrams in three segments, with evidence spanning position nodes.","gt_lineage_depth":4,"involved_gt_edges":["planar_distance","argmax"],"involved_gt_nodes":["ordered_frames","sensor_positions"]},"invo... | ||
What change occurred in the camera orientation from T1 to T2 in the top view? | [
"A. near reverse rotation",
"B. top view, rotate clockwise",
"C. basically maintain original direction",
"D. top view rotating counterclockwise"
] | D | single_choice | camera orientation change | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | medium | apartment_1_HAB_C06_Q01_HEADING_CHANGE_0129 | apartment_1:00045 | test | HAB_C06 | multi_view_choice | HAB_C06_Q01_HEADING_CHANGE | HAB_C06_Q01_HEADING_CHANGE_v05 | [
"A",
"B",
"C",
"D"
] | [
"near reverse rotation",
"top view, rotate clockwise",
"basically maintain original direction",
"top view rotating counterclockwise"
] | {"A":"near reverse rotation","B":"top view, rotate clockwise","C":"basically maintain original direction","D":"top view rotating counterclockwise"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | two time points with direction classes that are different. | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"turn_class":"top view rotating counterclockwise"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"two time points with direction classes that are different.","gt_lineage_depth":5,"involved_gt_edges":["forward_vector","signed_angle_delta","turn_class"],"involved_gt_nodes":["ordered_frames",... | ||
According to the top view on the right side, which direction is the camera T1 mainly facing? | [
"A. East-facing",
"B. West",
"C. south-facing",
"D. north-facing"
] | D | single_choice | camera orientation change | Multiple Choice Question | simple | apartment_1_HAB_C06_Q02_CARDINAL_HEADING_0001 | apartment_1:00256 | test | HAB_C06 | single_choice | HAB_C06_Q02_CARDINAL_HEADING | HAB_C06_Q02_CARDINAL_HEADING_v01 | [
"A",
"B",
"C",
"D"
] | [
"East-facing",
"West",
"south-facing",
"north-facing"
] | {"A":"East-facing","B":"West","C":"south-facing","D":"north-facing"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single time step pose field to direction category | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"cardinal_heading":"north-facing"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single time step pose field to direction category","gt_lineage_depth":4,"involved_gt_edges":["forward_vector","cardinal_bin"],"involved_gt_nodes":["frame","sensor_rotation"]},"involved_objects":[],"provenance... | ||
Which of these cells A to I appears to be more frontward? | [
"A. Middle-left cell",
"B. middle lower slot",
"C. Middle-right cell",
"D. Lower-right cell",
"E. left bottom cell",
"F. left top cell",
"G. middle upper slot",
"H. Upper-right cell",
"I. Center cell"
] | D | single_choice | image plane region localization | Location Selection Question | medium | apartment_1_HAB_C07_Q01_GRID_NEAREST_CELL_0003 | apartment_1:00186 | test | HAB_C07 | grounding_choice | HAB_C07_Q01_GRID_NEAREST_CELL | HAB_C07_Q01_GRID_NEAREST_CELL_v03 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"Middle-left cell",
"middle lower slot",
"Middle-right cell",
"Lower-right cell",
"left bottom cell",
"left top cell",
"middle upper slot",
"Upper-right cell",
"Center cell"
] | {"A":"Middle-left cell","B":"middle lower slot","C":"Middle-right cell","D":"Lower-right cell","E":"left bottom cell","F":"left top cell","G":"middle upper slot","H":"Upper-right cell","I":"Center cell"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statistics. | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":2.552,"B":3.05043,"C":2.6643,"D":1.76593,"E":2.13672,"F":2.23194,"G":2.5835,"H":2.46525,"I":3.32021}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statistics.... | ||
Which labeled area in the image has the most obvious distant view? | [
"A. right region",
"B. left area",
"C. central area"
] | C | single_choice | image plane region localization | Multiple Choice Question | medium | apartment_1_HAB_C07_Q02_FAR_SPACE_COLUMN_0008 | apartment_1:00049 | test | HAB_C07 | single_choice | HAB_C07_Q02_FAR_SPACE_COLUMN | HAB_C07_Q02_FAR_SPACE_COLUMN_v05 | [
"A",
"B",
"C"
] | [
"right region",
"left area",
"central area"
] | {"A":"right region","B":"left area","C":"central area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame region localization and far-field proportion statistics | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":0.0,"B":0.0,"C":0.11642}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame region localization and far-field proportion statistics","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["frame","depth_pixels","region_s... | ||
Order from nearest to farthest based on the planar distance from the subsequent acquisition points to T1. | [
"A. Time T3",
"B. Time T2",
"C. Time T4"
] | ["A","B","C"] | ordered_list | order of collection and multi-view consistency | Sorting Question | hard | apartment_1_HAB_C08_Q03_DISTANCE_FROM_START_ORDER_0001 | apartment_1:00213 | test | HAB_C08 | ordered_list | HAB_C08_Q03_DISTANCE_FROM_START_ORDER | HAB_C08_Q03_DISTANCE_FROM_START_ORDER_v02 | [
"A",
"B",
"C"
] | [
"Time T3",
"Time T2",
"Time T4"
] | {"A":"Time T3","B":"Time T2","C":"Time T4"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes. | [
"records.frame_index",
"records.sensor_states.depth_sensor.position"
] | {"answer_evidence":{"ordered_by_distance_from_T1":["A","B","C"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes.","gt_lineage_depth":4,"involved_gt_edges":["distance_to_reference","sort_ascending"],"involved... | ||
Which of the four labeled areas appears to be the closest to the camera? | [
"A. right lower area",
"B. left upper area",
"C. left bottom area",
"D. right upper area"
] | C | single_choice | single-frame regional depth localization | Location Selection Question | simple | apartment_1_HAB_C01_Q01_NEAREST_QUADRANT_0023 | apartment_1:00223 | test | HAB_C01 | grounding_choice | HAB_C01_Q01_NEAREST_QUADRANT | HAB_C01_Q01_NEAREST_QUADRANT_v02 | [
"A",
"B",
"C",
"D"
] | [
"right lower area",
"left upper area",
"left bottom area",
"right upper area"
] | {"A":"right lower area","B":"left upper area","C":"left bottom area","D":"right upper area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single frame, single field depth statistics, four-out-of-one region | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":3.2871,"B":2.81407,"C":2.44312,"D":3.60632}},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single frame, single field depth statistics, four-out-of-one region","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmin"],"involved_gt_nodes":["frame","depth... | ||
Compare the distances of the three areas in the image and give the order from nearest to farthest. | [
"A. left area",
"B. central area",
"C. right region"
] | ["B","C","A"] | ordered_list | single-frame regional depth localization | Sorting Question | medium | apartment_1_HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_0008 | apartment_1:00097 | test | HAB_C01 | ordered_list | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_v03 | [
"A",
"B",
"C"
] | [
"left area",
"central area",
"right region"
] | {"A":"left area","B":"central area","C":"right region"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Single frame with three object regions sorted, evidence chain includes three depth statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_near_to_far":["B","C","A"]},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Single frame with three object regions sorted, evidence chain includes three depth statistics.","gt_lineage_depth":5,"involved_gt_edges":["median","sort_ascending"],"involved_gt_nodes":["frame","depth_pixel... | ||
Which labeled area mainly consists of content close to the camera? | [
"A. left bottom area",
"B. left upper area",
"C. right lower area",
"D. right upper area"
] | B | single_choice | near-field spatial coverage | Location Selection Question | medium | apartment_1_HAB_C02_Q01_NEAR_SPACE_QUADRANT_0021 | apartment_1:00064 | test | HAB_C02 | grounding_choice | HAB_C02_Q01_NEAR_SPACE_QUADRANT | HAB_C02_Q01_NEAR_SPACE_QUADRANT_v04 | [
"A",
"B",
"C",
"D"
] | [
"left bottom area",
"left upper area",
"right lower area",
"right upper area"
] | {"A":"left bottom area","B":"left upper area","C":"right lower area","D":"right upper area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare the region sets by measuring the proportion of each covered by nearby content. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.87459,"B":0.91582,"C":0.16522,"D":0.4253}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Compare the region sets by measuring the proportion of each covered by nearby content.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nod... | ||
Arrange the annotated areas based on the amount of content close to the camera. | [
"A. right upper area",
"B. left upper area",
"C. left bottom area",
"D. right lower area"
] | ["A","D","B","C"] | ordered_list | near-field spatial coverage | Sorting Question | hard | apartment_1_HAB_C02_Q02_NEAR_SPACE_ORDER_0007 | apartment_1:00310 | test | HAB_C02 | ordered_list | HAB_C02_Q02_NEAR_SPACE_ORDER | HAB_C02_Q02_NEAR_SPACE_ORDER_v04 | [
"A",
"B",
"C",
"D"
] | [
"right upper area",
"left upper area",
"left bottom area",
"right lower area"
] | {"A":"right upper area","B":"left upper area","C":"left bottom area","D":"right lower area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Rank four region sets in a single frame using four measurements of nearby coverage. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_high_to_low_near_fraction":["A","D","B","C"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Rank four region sets in a single frame using four measurements of nearby coverage.","gt_lineage_depth":5,"involved_gt_edges":["near_threshold_fraction","sort_descending"],"involved_gt_nodes... | ||
Which area appears to have the most depth? | [
"A. right region",
"B. left area",
"C. central area"
] | A | single_choice | depth undulation and transparency | Multiple Choice Question | medium | apartment_1_HAB_C03_Q01_DEPTH_VARIATION_REGION_0037 | apartment_1:00014 | test | HAB_C03 | single_choice | HAB_C03_Q01_DEPTH_VARIATION_REGION | HAB_C03_Q01_DEPTH_VARIATION_REGION_v04 | [
"A",
"B",
"C"
] | [
"right region",
"left area",
"central area"
] | {"A":"right region","B":"left area","C":"central area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Comparing regional geometric variations in a single frame requires quantile statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":1.97478,"B":0.34073,"C":0.61799}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Comparing regional geometric variations in a single frame requires quantile statistics.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["fram... | ||
On the top-down trajectory at right, which segment between consecutive capture points covers the greatest distance? | [
"A. T2→T3",
"B. T1→T2",
"C. T3→T4"
] | B | single_choice | camera plane trajectory | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | hard | apartment_1_HAB_C05_Q01_LARGEST_STEP_0014 | apartment_1:00060 | test | HAB_C05 | multi_view_choice | HAB_C05_Q01_LARGEST_STEP | HAB_C05_Q01_LARGEST_STEP_v01 | [
"A",
"B",
"C"
] | [
"T2→T3",
"T1→T2",
"T3→T4"
] | {"A":"T2→T3","B":"T1→T2","C":"T3→T4"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare trajectory diagrams in three segments, with evidence spanning position nodes. | [
"records.sensor_states.depth_sensor.position",
"records.frame_index"
] | {"answer_evidence":{"largest_step":"T1→T2"},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Compare trajectory diagrams in three segments, with evidence spanning position nodes.","gt_lineage_depth":4,"involved_gt_edges":["planar_distance","argmax"],"involved_gt_nodes":["ordered_frames","sensor_positions"]},"invo... | ||
Look at the top-down arrow on the right. Which option best matches the direction change from T1 to T2? | [
"A. basically maintain original direction",
"B. top view, rotate clockwise",
"C. top view rotating counterclockwise",
"D. near reverse rotation"
] | D | single_choice | camera orientation change | Multiple Image/Multiple Viewpoint Consistency Multiple Choice Question | medium | apartment_1_HAB_C06_Q01_HEADING_CHANGE_0205 | apartment_1:00283 | test | HAB_C06 | multi_view_choice | HAB_C06_Q01_HEADING_CHANGE | HAB_C06_Q01_HEADING_CHANGE_v01 | [
"A",
"B",
"C",
"D"
] | [
"basically maintain original direction",
"top view, rotate clockwise",
"top view rotating counterclockwise",
"near reverse rotation"
] | {"A":"basically maintain original direction","B":"top view, rotate clockwise","C":"top view rotating counterclockwise","D":"near reverse rotation"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | two time points with direction classes that are different. | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"turn_class":"near reverse rotation"},"gt_lineage":{"difficulty":"medium","difficulty_reason":"two time points with direction classes that are different.","gt_lineage_depth":5,"involved_gt_edges":["forward_vector","signed_angle_delta","turn_class"],"involved_gt_nodes":["ordered_frames","sensor_rotat... | ||
Look at the arrow pointing to the direction of T1, which main direction is it closer to? | [
"A. West",
"B. north-facing",
"C. south-facing",
"D. East-facing"
] | A | single_choice | camera orientation change | Multiple Choice Question | simple | apartment_1_HAB_C06_Q02_CARDINAL_HEADING_0002 | apartment_1:00265 | test | HAB_C06 | single_choice | HAB_C06_Q02_CARDINAL_HEADING | HAB_C06_Q02_CARDINAL_HEADING_v02 | [
"A",
"B",
"C",
"D"
] | [
"West",
"north-facing",
"south-facing",
"East-facing"
] | {"A":"West","B":"north-facing","C":"south-facing","D":"East-facing"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single time step pose field to direction category | [
"records.sensor_states.depth_sensor.rotation",
"records.frame_index"
] | {"answer_evidence":{"cardinal_heading":"West"},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single time step pose field to direction category","gt_lineage_depth":4,"involved_gt_edges":["forward_vector","cardinal_bin"],"involved_gt_nodes":["frame","sensor_rotation"]},"involved_objects":[],"provenance":{"gt_f... | ||
Look at the annotated image and select the cell closest to the camera. | [
"A. Lower-right cell",
"B. middle upper slot",
"C. Center cell",
"D. left bottom cell",
"E. Upper-right cell",
"F. Middle-left cell",
"G. Middle-right cell",
"H. middle lower slot",
"I. left top cell"
] | A | single_choice | image plane region localization | Location Selection Question | medium | apartment_1_HAB_C07_Q01_GRID_NEAREST_CELL_0017 | apartment_1:00211 | test | HAB_C07 | grounding_choice | HAB_C07_Q01_GRID_NEAREST_CELL | HAB_C07_Q01_GRID_NEAREST_CELL_v02 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H",
"I"
] | [
"Lower-right cell",
"middle upper slot",
"Center cell",
"left bottom cell",
"Upper-right cell",
"Middle-left cell",
"Middle-right cell",
"middle lower slot",
"left top cell"
] | {"A":"Lower-right cell","B":"middle upper slot","C":"Center cell","D":"left bottom cell","E":"Upper-right cell","F":"Middle-left cell","G":"Middle-right cell","H":"middle lower slot","I":"left top cell"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statistics. | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":1.55444,"B":2.62707,"C":3.05287,"D":2.49393,"E":2.65766,"F":2.50568,"G":3.9266,"H":2.08743,"I":2.34958}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame nine-region localization selection, with more candidates but still uniquely determined by deep statisti... | ||
Which of the three regions A, B, and C appears to be farthest away? | [
"A. right region",
"B. central area",
"C. left area"
] | B | single_choice | image plane region localization | Multiple Choice Question | medium | apartment_1_HAB_C07_Q02_FAR_SPACE_COLUMN_0014 | apartment_1:00202 | test | HAB_C07 | single_choice | HAB_C07_Q02_FAR_SPACE_COLUMN | HAB_C07_Q02_FAR_SPACE_COLUMN_v01 | [
"A",
"B",
"C"
] | [
"right region",
"central area",
"left area"
] | {"A":"right region","B":"central area","C":"left area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single-frame region localization and far-field proportion statistics | [
"records.rgb_path",
"records.depth_path"
] | {"answer_evidence":{"value_by_label":{"A":0.0,"B":0.04063,"C":0.0}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"single-frame region localization and far-field proportion statistics","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_nodes":["frame","depth_pixels","region_s... | ||
View the top-down trajectory on the right side. Please sort T2, T3, T4 from closest to farthest from T1. | [
"A. Time T4",
"B. Time T2",
"C. Time T3"
] | ["B","A","C"] | ordered_list | order of collection and multi-view consistency | Sorting Question | hard | apartment_1_HAB_C08_Q03_DISTANCE_FROM_START_ORDER_0005 | apartment_1:00326 | test | HAB_C08 | ordered_list | HAB_C08_Q03_DISTANCE_FROM_START_ORDER | HAB_C08_Q03_DISTANCE_FROM_START_ORDER_v01 | [
"A",
"B",
"C"
] | [
"Time T4",
"Time T2",
"Time T3"
] | {"A":"Time T4","B":"Time T2","C":"Time T3"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes. | [
"records.frame_index",
"records.sensor_states.depth_sensor.position"
] | {"answer_evidence":{"ordered_by_distance_from_T1":["B","A","C"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Multi-candidate trajectory distance ranking referenced to T1, with evidence spanning multiple pose nodes.","gt_lineage_depth":4,"involved_gt_edges":["distance_to_reference","sort_ascending"],"involved... | ||
Which of the four labeled areas appears to be the closest to the camera? | [
"A. left upper area",
"B. right lower area",
"C. right upper area",
"D. left bottom area"
] | B | single_choice | single-frame regional depth localization | Location Selection Question | simple | apartment_1_HAB_C01_Q01_NEAREST_QUADRANT_0033 | apartment_1:00324 | test | HAB_C01 | grounding_choice | HAB_C01_Q01_NEAREST_QUADRANT | HAB_C01_Q01_NEAREST_QUADRANT_v02 | [
"A",
"B",
"C",
"D"
] | [
"left upper area",
"right lower area",
"right upper area",
"left bottom area"
] | {"A":"left upper area","B":"right lower area","C":"right upper area","D":"left bottom area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | single frame, single field depth statistics, four-out-of-one region | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":3.07095,"B":2.46189,"C":2.70405,"D":3.21248}},"gt_lineage":{"difficulty":"simple","difficulty_reason":"single frame, single field depth statistics, four-out-of-one region","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmin"],"involved_gt_nodes":["frame","dept... | ||
Arrange A, B, and C according to their spatial distance in the picture. | [
"A. left area",
"B. central area",
"C. right region"
] | ["A","B","C"] | ordered_list | single-frame regional depth localization | Sorting Question | medium | apartment_1_HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_0019 | apartment_1:00286 | test | HAB_C01 | ordered_list | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER | HAB_C01_Q03_HORIZONTAL_DEPTH_ORDER_v04 | [
"A",
"B",
"C"
] | [
"left area",
"central area",
"right region"
] | {"A":"left area","B":"central area","C":"right region"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Single frame with three object regions sorted, evidence chain includes three depth statistics. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_near_to_far":["A","B","C"]},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Single frame with three object regions sorted, evidence chain includes three depth statistics.","gt_lineage_depth":5,"involved_gt_edges":["median","sort_ascending"],"involved_gt_nodes":["frame","depth_pixel... | ||
Which annotated area has the most nearby content? | [
"A. left bottom area",
"B. right lower area",
"C. left upper area",
"D. right upper area"
] | D | single_choice | near-field spatial coverage | Location Selection Question | medium | apartment_1_HAB_C02_Q01_NEAR_SPACE_QUADRANT_0023 | apartment_1:00136 | test | HAB_C02 | grounding_choice | HAB_C02_Q01_NEAR_SPACE_QUADRANT | HAB_C02_Q01_NEAR_SPACE_QUADRANT_v01 | [
"A",
"B",
"C",
"D"
] | [
"left bottom area",
"right lower area",
"left upper area",
"right upper area"
] | {"A":"left bottom area","B":"right lower area","C":"left upper area","D":"right upper area"} | scalar | Accuracy | choice_exact_match | habitat_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key, matching VLMEvalKit image_mcq exact_matching/report_acc behavior. | true | Compare the region sets by measuring the proportion of each covered by nearby content. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"value_by_label":{"A":0.23542,"B":0.61888,"C":0.13871,"D":0.66562}},"gt_lineage":{"difficulty":"medium","difficulty_reason":"Compare the region sets by measuring the proportion of each covered by nearby content.","gt_lineage_depth":5,"involved_gt_edges":["region_depth_stat","argmax"],"involved_gt_no... | ||
Which areas are more like the foreground? Please rank them from highest to lowest degree of resemblance. | [
"A. left upper area",
"B. left bottom area",
"C. right upper area",
"D. right lower area"
] | ["C","D","A","B"] | ordered_list | near-field spatial coverage | Sorting Question | hard | apartment_1_HAB_C02_Q02_NEAR_SPACE_ORDER_0010 | apartment_1:00103 | test | HAB_C02 | ordered_list | HAB_C02_Q02_NEAR_SPACE_ORDER | HAB_C02_Q02_NEAR_SPACE_ORDER_v02 | [
"A",
"B",
"C",
"D"
] | [
"left upper area",
"left bottom area",
"right upper area",
"right lower area"
] | {"A":"left upper area","B":"left bottom area","C":"right upper area","D":"right lower area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | habitat_gt_objective_metrics.ordered_scores | Exact full permutation plus fraction of pairwise order relations that match GT. | true | Rank four region sets in a single frame using four measurements of nearby coverage. | [
"records.depth_path",
"records.rgb_path"
] | {"answer_evidence":{"ordered_high_to_low_near_fraction":["C","D","A","B"]},"gt_lineage":{"difficulty":"hard","difficulty_reason":"Rank four region sets in a single frame using four measurements of nearby coverage.","gt_lineage_depth":5,"involved_gt_edges":["near_threshold_fraction","sort_descending"],"involved_gt_nodes... |
End of preview. Expand in Data Studio
BenchForge Habitat
Indoor navigation and reasoning over Habitat views and trajectories.
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, andanswer: 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-habitat", 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
- 82