Upload folder using huggingface_hub

.gitattributes

@@ -52,3 +52,173 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_accepted.png filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_intended.png filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_measured.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/detect_refine_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/detect_refine_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/detect_refine_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/diversify_input_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/diversify_input_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/crops/diversify_input_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/references/ref_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/references/ref_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000001/references/ref_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/detect_refine_city_buildings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/detect_refine_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/detect_refine_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/detect_refine_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/detect_refine_vehicle_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/diversify_input_city_buildings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/diversify_input_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/diversify_input_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/diversify_input_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/crops/diversify_input_vehicle_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_city_buildings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_drainage_grate.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_parked_car_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_parked_suv_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_pedestrian_walking.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_shop_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_storefront_sign.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_street_signs.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_traveling_dark_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_vehicle_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_white_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000002/references/ref_yellow_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/detect_refine_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/detect_refine_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/detect_refine_waiting_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/diversify_input_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/diversify_input_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/crops/diversify_input_waiting_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/references/ref_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/references/ref_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000003/references/ref_waiting_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/crops/detect_refine_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/crops/detect_refine_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/crops/diversify_input_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/crops/diversify_input_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/references/ref_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/references/ref_delivery_truck.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/references/ref_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/references/ref_traffic_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000004/references/ref_walker.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/crops/detect_refine_concrete_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/crops/detect_refine_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/crops/diversify_input_concrete_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/crops/diversify_input_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/crops/diversify_input_uniformed_officer.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_bystander_in_suit.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_concrete_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_firefighter.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_traffic_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000005/references/ref_uniformed_officer.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/crops/detect_refine_dark_building_facade.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/crops/detect_refine_double_solid_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/crops/diversify_input_dark_building_facade.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/crops/diversify_input_double_solid_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/crops/diversify_input_white_panel_van.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_awning_building_corner.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_dark_building_facade.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_distant_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_double_solid_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_emergency_vehicle.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_green_street_sign.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_vertical_illuminated_sign.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000006/references/ref_white_panel_van.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_brick_building_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_crosswalk_markings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_iron_balcony.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_multi_story_building_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_overhead_wires.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/detect_refine_street_light_pole.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/diversify_input_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/diversify_input_iron_balcony.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/diversify_input_multi_story_building_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/diversify_input_overhead_wires.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/crops/diversify_input_street_light_pole.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_brick_building_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_bunch_of_balloons.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_crosswalk_markings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_dark_car_1.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_dark_car_2.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_iron_balcony.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_multi_story_building_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_overhead_wires.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_white_garbage_bag.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_white_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000007/references/ref_woman_in_dark_dress.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/detect_refine_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/detect_refine_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/detect_refine_street_lamp.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_sign_holder.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_street_lamp.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_woman_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/crops/diversify_input_young_man_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_businessman_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_pedestrian_crossing_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_pedestrian_standing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_pedestrian_walking_away_sidewalk.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_sign_holder.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_street_lamp.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_woman_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000008/references/ref_young_man_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/detect_refine_metal_railing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/detect_refine_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/detect_refine_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/diversify_input_metal_railing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/diversify_input_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/diversify_input_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/crops/diversify_input_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/references/ref_metal_railing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/references/ref_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/references/ref_pedestrian_in_suit.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000009/references/ref_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/detect_refine_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/detect_refine_man_in_grey_sweater.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/detect_refine_pedestrian_in_light_blue.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/detect_refine_pedestrian_in_striped_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/detect_refine_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_man_in_grey_sweater.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_man_in_pink_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_pedestrian_in_light_blue.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_pedestrian_in_striped_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/crops/diversify_input_pedestrian_with_backpack.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_man_in_grey_sweater.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_man_in_pink_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_in_light_blue.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_in_light_jacket.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_in_red.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_in_striped_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/sample_000010/references/ref_pedestrian_with_backpack.png filter=lfs diff=lfs merge=lfs -text

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/README.md

@@ -0,0 +1,16 @@
+# samples_v8
+
+Generated with `data_recipe_v8.md`: vocabulary-first planning, adaptive canvas selection, structured JSON compose prompts, no identity verification, no gate, SAM white-background reference postprocessing, and strict reference-completeness verification with regenerate-until-pass behavior.
+
+- chat model: `gcp/google/gemini-3.1-pro-preview`
+- image model: `gcp/google/gemini-3-pro-image-preview`
+- people references: `white_bg_full_body_front`
+- non-person references: `white_bg_encyclopedia_photo`
+- SAM postprocess: every generated reference is segmented with `sam_vit_b` and pasted onto pure `#ffffff` background
+- reference verify max attempts per subject: `10`
+- allowed canvases: `[{"aspect_ratio": "1:1", "size": [1024, 1024], "style": "photorealistic"}, {"aspect_ratio": "4:3", "size": [1152, 864], "style": "photorealistic"}, {"aspect_ratio": "3:4", "size": [864, 1152], "style": "photorealistic"}, {"aspect_ratio": "3:2", "size": [1248, 832], "style": "photorealistic"}, {"aspect_ratio": "2:3", "size": [832, 1248], "style": "photorealistic"}, {"aspect_ratio": "16:9", "size": [1280, 720], "style": "photorealistic"}, {"aspect_ratio": "9:16", "size": [720, 1280], "style": "photorealistic"}]`
+- scenario mode: `driving`
+- pools: `vocab_task_pool`, `plan_pool`, `scene_pool`, `detection_pool`, `reference_pool`
+- bbox overlay: `bbox_overlay.png` draws every planned subject bbox; a sample is rejected and regenerated if any planned subject is still missing after VLM detection retries
+- detection max attempts per subject: `3`
+- launch args: `{"compose_workers": 3, "detect_max_attempts": 3, "detect_workers": 3, "emit_workers": 4, "idle_sleep": 1.0, "image_inflight": 32, "image_interval": 0.05, "image_max_retries": 8, "max_retries": 3, "no_topup": false, "plan_workers": 6, "ref_verify_max_attempts": 10, "reference_workers": 6, "requeue_in_progress": true, "seed": 1781927993, "status_interval": 30.0, "subject_detect_workers": 24, "target_samples": 10}`

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000001.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "main_image": "sample_000001/main_image.png",
+  "detections": "sample_000001/detections.json",
+  "n_detected": 3,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000002.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "main_image": "sample_000002/main_image.png",
+  "detections": "sample_000002/detections.json",
+  "n_detected": 15,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000003.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "main_image": "sample_000003/main_image.png",
+  "detections": "sample_000003/detections.json",
+  "n_detected": 3,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000004.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "main_image": "sample_000004/main_image.png",
+  "detections": "sample_000004/detections.json",
+  "n_detected": 5,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000005.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "main_image": "sample_000005/main_image.png",
+  "detections": "sample_000005/detections.json",
+  "n_detected": 6,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000006.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "main_image": "sample_000006/main_image.png",
+  "detections": "sample_000006/detections.json",
+  "n_detected": 8,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000007.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000007",
+  "plan_path": "sample_000007/plan.json",
+  "task_path": "sample_000007/vocab_task.json",
+  "main_image": "sample_000007/main_image.png",
+  "detections": "sample_000007/detections.json",
+  "n_detected": 14,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000007",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000008.json

@@ -0,0 +1,22 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "main_image": "sample_000008/main_image.png",
+  "detections": "sample_000008/detections.json",
+  "n_detected": 10,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "detection_pool",
+  "retry_count": 1,
+  "errors": [
+    {
+      "time": 1782293479.3988369,
+      "error": "RuntimeError: reference generation or verification failed for street_lamp: RuntimeError: reference verification failed for street_lamp after 10 attempts: heavily truncated; missing main identifying part (lamp fixture)",
+      "traceback": "Traceback (most recent call last):\n  File \"/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/samples_v8/generate_samples_v8.py\", line 866, in generate_references\n    references.append(helpers.diversify_subject(get_client(), sdir, main_image, subject, detection))\n                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/samples_v8/v8_helpers.py\", line 693, in diversify_subject\n    raise RuntimeError(\nRuntimeError: reference verification failed for street_lamp after 10 attempts: heavily truncated; missing main identifying part (lamp fixture)\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n  File \"/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/samples_v8/generate_samples_v8.py\", line 1020, in worker_loop\n    handler(manifest)\n  File \"/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/samples_v8/generate_samples_v8.py\", line 1113, in handler\n    references, reference_errors = generate_references(sample_id, plan, detections)\n                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/samples_v8/generate_samples_v8.py\", line 871, in generate_references\n    raise RuntimeError(f\"reference generation or verification failed for {name}: {errors[name]}\") from exc\nRuntimeError: reference generation or verification failed for street_lamp: RuntimeError: reference verification failed for street_lamp after 10 attempts: heavily truncated; missing main identifying part (lamp fixture)\n"
+    }
+  ]
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000009.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "main_image": "sample_000009/main_image.png",
+  "detections": "sample_000009/detections.json",
+  "n_detected": 5,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/detection_pool/done/sample_000010.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "main_image": "sample_000010/main_image.png",
+  "detections": "sample_000010/detections.json",
+  "n_detected": 9,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000001.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000001",
+  "row": "sample_000001/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000002.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000002",
+  "row": "sample_000002/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000003.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000003",
+  "row": "sample_000003/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000004.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000004",
+  "row": "sample_000004/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000005.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000005",
+  "row": "sample_000005/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000006.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000006",
+  "row": "sample_000006/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000007.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000007",
+  "row": "sample_000007/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000008.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000008",
+  "row": "sample_000008/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000009.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000009",
+  "row": "sample_000009/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/emit_pool/done/sample_000010.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000010",
+  "row": "sample_000010/row.json"
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000001.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "prompt_hash": "f859db774945f5b6373454badaca82233955ab8ca8b947dbecdab4a6975bb976",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000002.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "prompt_hash": "0e4f0ac71145112a7a163e8d8798062138922bb63e746f904a784c8608c2d011",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000003.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "prompt_hash": "325b286ae7c454ccf5a9081711812c3439116af9f5e2b39f2029ac6c415ac47e",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000004.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "prompt_hash": "9d98080d776626e886894383e176082ba23f7c536f920f60fd7462fb1e21e246",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000005.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "prompt_hash": "dd17f99d7265de9375dbd703b97e688634f8632997dd81e3c11f787d9cb95361",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000006.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "prompt_hash": "98e81e2eedabb51a9bd1b9a35847125d84c5ef899ddbf672c226fe62b69506b1",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000007.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000007",
+  "plan_path": "sample_000007/plan.json",
+  "task_path": "sample_000007/vocab_task.json",
+  "prompt_hash": "a11d77c30b9e5c0a4d946297ad0615803424cd56063dca0351a861c7c079059b",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000007",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000008.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "prompt_hash": "fd36686545adb806e34c1b62a7004d06a56e505d47d056ee5b0b89752f3d75bd",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000009.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "prompt_hash": "bddb7951d32d26de33b4ea1a6aa00cdbd7da12d9349508f23b30d7e3e1eadef8",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/plan_pool/done/sample_000010.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "prompt_hash": "cc6385c75e6510ab98831d1b4008bb6fc0bdaf521d9d020a82a36d3a08b9d8d8",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000001.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "main_image": "sample_000001/main_image.png",
+  "detections": "sample_000001/detections.json",
+  "references": "sample_000001/references.json",
+  "n_references": 3,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000002.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "main_image": "sample_000002/main_image.png",
+  "detections": "sample_000002/detections.json",
+  "references": "sample_000002/references.json",
+  "n_references": 15,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000003.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "main_image": "sample_000003/main_image.png",
+  "detections": "sample_000003/detections.json",
+  "references": "sample_000003/references.json",
+  "n_references": 3,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000004.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "main_image": "sample_000004/main_image.png",
+  "detections": "sample_000004/detections.json",
+  "references": "sample_000004/references.json",
+  "n_references": 5,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000005.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "main_image": "sample_000005/main_image.png",
+  "detections": "sample_000005/detections.json",
+  "references": "sample_000005/references.json",
+  "n_references": 6,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000006.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "main_image": "sample_000006/main_image.png",
+  "detections": "sample_000006/detections.json",
+  "references": "sample_000006/references.json",
+  "n_references": 8,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000007.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000007",
+  "plan_path": "sample_000007/plan.json",
+  "task_path": "sample_000007/vocab_task.json",
+  "main_image": "sample_000007/main_image.png",
+  "detections": "sample_000007/detections.json",
+  "references": "sample_000007/references.json",
+  "n_references": 14,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000007",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000008.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "main_image": "sample_000008/main_image.png",
+  "detections": "sample_000008/detections.json",
+  "references": "sample_000008/references.json",
+  "n_references": 10,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000009.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "main_image": "sample_000009/main_image.png",
+  "detections": "sample_000009/detections.json",
+  "references": "sample_000009/references.json",
+  "n_references": 5,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/reference_pool/done/sample_000010.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "main_image": "sample_000010/main_image.png",
+  "detections": "sample_000010/detections.json",
+  "references": "sample_000010/references.json",
+  "n_references": 9,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000001.json

@@ -0,0 +1,164 @@
+{
+  "sample_id": "sample_000001",
+  "target_total": 3,
+  "target_people": 1,
+  "target_objects": 2,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 3,
+  "n_detected": 3,
+  "n_subjects": 3,
+  "subjects": [
+    {
+      "name": "pedestrian",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_69/273278,12fc4700013112375.jpg:person:3",
+      "source_name": "pedestrian",
+      "source_description": "A person wearing a dark coat and trousers. Source dataset: CrowdHuman. Scene context: A bustling city street lined with trees showcasing vibrant yellow autumn foliage, with many pedestrians walking in both directions.",
+      "sub_caption": "pedestrian: A person wearing a dark coat and trousers, walking confidently.. Scene role: Walking along the curbside near the barrier.",
+      "measured_bbox": [
+        0.1528,
+        0.301,
+        0.2511,
+        0.7071
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_pedestrian_attempt_01.png",
+        "output": "references/ref_pedestrian.png",
+        "mask": "references/sam_mask_pedestrian.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          330.0,
+          42.0,
+          698.0,
+          1007.0
+        ],
+        "mask_score": 3.413244,
+        "mask_area_ratio": 0.159381,
+        "elapsed_seconds": 33.2771
+      }
+    },
+    {
+      "name": "parked_dark_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bcb356f6-520dd65c:object:9",
+      "source_name": "parked dark car",
+      "source_description": "A dark-colored sedan parked on the right side of the street, behind the silver car. Source dataset: BDD100K. Scene context: A large white New York City bus is driving down a city street on an overcast day with other cars parked and driving around it.",
+      "sub_caption": "parked dark car: A dark-colored sedan.. Scene role: Parked on the street near the curb in the background right.",
+      "measured_bbox": [
+        0.5163,
+        0.3897,
+        0.9968,
+        0.9244
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_parked_dark_car.png",
+      "raw_ref_image": "references/raw_ref_parked_dark_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_parked_dark_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_parked_dark_car_attempt_01.png",
+        "output": "references/ref_parked_dark_car.png",
+        "mask": "references/sam_mask_parked_dark_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          47.0,
+          315.0,
+          976.0,
+          694.0
+        ],
+        "mask_score": 3.4345,
+        "mask_area_ratio": 0.180014,
+        "elapsed_seconds": 7.1991
+      }
+    },
+    {
+      "name": "metal_barrier",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_74/284193,1da20000b642be5b.jpg:object:5",
+      "source_name": "metal barrier",
+      "source_description": "silver metal barricade placed near the entrance Source dataset: CrowdHuman. Scene context: People are gathered outside the entrance of a stone building with arched doorways and large windows.",
+      "sub_caption": "metal barrier: A silver metal barricade placed along the street.. Scene role: Positioned along the curb to section off the pedestrian area from the road.",
+      "measured_bbox": [
+        0.3454,
+        0.4302,
+        0.5465,
+        0.8402
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_metal_barrier.png",
+      "raw_ref_image": "references/raw_ref_metal_barrier_attempt_01.png",
+      "reference_verify": "references/reference_verify_metal_barrier.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_metal_barrier_attempt_01.png",
+        "output": "references/ref_metal_barrier.png",
+        "mask": "references/sam_mask_metal_barrier.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          2.0,
+          107.0,
+          1009.0,
+          986.0
+        ],
+        "mask_score": 1.555076,
+        "mask_area_ratio": 0.845579,
+        "elapsed_seconds": 7.2854
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000002.json

@@ -0,0 +1,716 @@
+{
+  "sample_id": "sample_000002",
+  "target_total": 15,
+  "target_people": 2,
+  "target_objects": 13,
+  "canvas_size": [
+    1152,
+    864
+  ],
+  "canvas_aspect_ratio": "4:3",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 15,
+  "n_detected": 15,
+  "n_subjects": 15,
+  "subjects": [
+    {
+      "name": "pedestrian_walking",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "BDD100K:c5d864fa-b0b2380b:person:2",
+      "source_name": "pedestrian",
+      "source_description": "Person walking away from the camera on the right sidewalk, wearing dark clothing. Source dataset: BDD100K. Scene context: View from inside a vehicle driving down a wet city street lined with tall buildings, with pedestrians on the sidewalk and several cars and taxis ahead in the rain.",
+      "sub_caption": "pedestrian: Person walking away from the camera on the right sidewalk, wearing dark clothing.. Scene role: walking along the sidewalk on the right side of the street",
+      "measured_bbox": [
+        0.7497,
+        0.4757,
+        0.7954,
+        0.6192
+      ],
+      "detection_confidence": 100,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian_walking.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_walking_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian_walking.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_pedestrian_walking_attempt_01.png",
+        "output": "references/ref_pedestrian_walking.png",
+        "mask": "references/sam_mask_pedestrian_walking.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          341.0,
+          63.0,
+          695.0,
+          972.0
+        ],
+        "mask_score": 3.459152,
+        "mask_area_ratio": 0.145545,
+        "elapsed_seconds": 8.3331
+      }
+    },
+    {
+      "name": "shop_pedestrian",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "BDD100K:b714a088-861a043b:person:2",
+      "source_name": "pedestrian",
+      "source_description": "another person near the shop entrance on the left, partially obscured Source dataset: BDD100K. Scene context: a city street scene from the perspective of a vehicle, showing multiple cars in motion, buildings lining the road, and pedestrians on the sidewalks during dusk or early evening",
+      "sub_caption": "pedestrian: Person standing near a shop entrance on the right, partially obscured.. Scene role: standing on the sidewalk near the storefronts on the right",
+      "measured_bbox": [
+        0.9337,
+        0.4752,
+        0.9695,
+        0.6107
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_shop_pedestrian.png",
+      "raw_ref_image": "references/raw_ref_shop_pedestrian_attempt_02.png",
+      "reference_verify": "references/reference_verify_shop_pedestrian.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 2,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_shop_pedestrian_attempt_02.png",
+        "output": "references/ref_shop_pedestrian.png",
+        "mask": "references/sam_mask_shop_pedestrian.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          312.0,
+          43.0,
+          719.0,
+          1020.0
+        ],
+        "mask_score": 3.162079,
+        "mask_area_ratio": 0.167512,
+        "elapsed_seconds": 7.2283
+      }
+    },
+    {
+      "name": "city_buildings",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_9/283554,31eeb000e9237b31.jpg:object:9",
+      "source_name": "building",
+      "source_description": "Various city buildings of different heights forming the skyline in the background. Source dataset: CrowdHuman. Scene context: People are walking and resting on the wooden walkway of a large suspension bridge with a city skyline in the background.",
+      "sub_caption": "building: Various city buildings of different heights forming the skyline and lining the street.. Scene role: framing the street and forming the background skyline",
+      "measured_bbox": [
+        0.0,
+        0.0,
+        1.0,
+        0.6084
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_city_buildings.png",
+      "raw_ref_image": "references/raw_ref_city_buildings_attempt_01.png",
+      "reference_verify": "references/reference_verify_city_buildings.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_city_buildings_attempt_01.png",
+        "output": "references/ref_city_buildings.png",
+        "mask": "references/sam_mask_city_buildings.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          14.0,
+          171.0,
+          1009.0,
+          883.0
+        ],
+        "mask_score": 3.176814,
+        "mask_area_ratio": 0.327415,
+        "elapsed_seconds": 7.146
+      }
+    },
+    {
+      "name": "street_signs",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b5ab0e46-8eab4733:object:8",
+      "source_name": "street signs",
+      "source_description": "Various street signs attached to a pole on the right side of the street, including a speed limit sign. Source dataset: BDD100K. Scene context: A narrow city street with several parked and moving box trucks and vans, bordered by multi-story buildings and construction barriers.",
+      "sub_caption": "street signs: Various blank street signs attached to a pole on the right side of the street.. Scene role: mounted on a pole next to the right sidewalk",
+      "measured_bbox": [
+        0.641,
+        0.165,
+        0.744,
+        0.408
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_signs.png",
+      "raw_ref_image": "references/raw_ref_street_signs_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_signs.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_signs_attempt_01.png",
+        "output": "references/ref_street_signs.png",
+        "mask": "references/sam_mask_street_signs.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          224.0,
+          0.0,
+          744.0,
+          1023.0
+        ],
+        "mask_score": 3.332549,
+        "mask_area_ratio": 0.190769,
+        "elapsed_seconds": 7.1886
+      }
+    },
+    {
+      "name": "storefront_sign",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_4/273275,46a6f0005d04fc24.jpg:object:7",
+      "source_name": "storefront sign",
+      "source_description": "A dark sign with white text 'DELI' and a smaller red 'ATM' sign beneath it on the left. Source dataset: CrowdHuman. Scene context: A densely crowded city street with numerous pedestrians walking towards the camera, while emergency vehicles with flashing lights are visible in the background traffic.",
+      "sub_caption": "storefront sign: A dark hanging sign framework attached to a building on the right, devoid of readable text.. Scene role: hanging above the shop entrance on the right side of the road",
+      "measured_bbox": [
+        0.7854,
+        0.1934,
+        0.9082,
+        0.2906
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_storefront_sign.png",
+      "raw_ref_image": "references/raw_ref_storefront_sign_attempt_01.png",
+      "reference_verify": "references/reference_verify_storefront_sign.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_storefront_sign_attempt_01.png",
+        "output": "references/ref_storefront_sign.png",
+        "mask": "references/sam_mask_storefront_sign.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          46.0,
+          0.0,
+          1023.0,
+          811.0
+        ],
+        "mask_score": 3.296373,
+        "mask_area_ratio": 0.447847,
+        "elapsed_seconds": 7.3102
+      }
+    },
+    {
+      "name": "parked_suv_right",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c54441e6-400c221e:object:4",
+      "source_name": "parked SUV",
+      "source_description": "Dark-colored SUV parked ahead of the sedan on the right side of the road. Source dataset: BDD100K. Scene context: Nighttime driving scene approaching an intersection with green traffic lights and parked cars on the right.",
+      "sub_caption": "parked SUV: Dark-colored SUV parked on the right side of the road.. Scene role: parked alongside the right curb",
+      "measured_bbox": [
+        0.5507,
+        0.4879,
+        0.6783,
+        0.6234
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_parked_suv_right.png",
+      "raw_ref_image": "references/raw_ref_parked_suv_right_attempt_01.png",
+      "reference_verify": "references/reference_verify_parked_suv_right.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_parked_suv_right_attempt_01.png",
+        "output": "references/ref_parked_suv_right.png",
+        "mask": "references/sam_mask_parked_suv_right.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          156.0,
+          150.0,
+          868.0,
+          812.0
+        ],
+        "mask_score": 3.463227,
+        "mask_area_ratio": 0.291222,
+        "elapsed_seconds": 7.2583
+      }
+    },
+    {
+      "name": "parked_car_left",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bb2e43e4-5e7a7129:object:6",
+      "source_name": "car",
+      "source_description": "A dark car parked along the left curb further ahead. Source dataset: BDD100K. Scene context: A dashcam view from a vehicle driving down a multi-lane road on a partly cloudy day, with a few other cars and pedestrians on the sidewalk.",
+      "sub_caption": "car: A dark car parked along the left curb further ahead.. Scene role: parked alongside the left curb",
+      "measured_bbox": [
+        0.0,
+        0.5102,
+        0.1259,
+        0.5998
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_parked_car_left.png",
+      "raw_ref_image": "references/raw_ref_parked_car_left_attempt_01.png",
+      "reference_verify": "references/reference_verify_parked_car_left.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_parked_car_left_attempt_01.png",
+        "output": "references/ref_parked_car_left.png",
+        "mask": "references/sam_mask_parked_car_left.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          319.0,
+          1023.0,
+          695.0
+        ],
+        "mask_score": 3.122119,
+        "mask_area_ratio": 0.19451,
+        "elapsed_seconds": 8.5738
+      }
+    },
+    {
+      "name": "traveling_dark_suv",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c889c950-865ca5b6:object:0",
+      "source_name": "dark SUV",
+      "source_description": "A dark SUV traveling in the left lane, with visible red taillights. Source dataset: BDD100K. Scene context: Nighttime driving on a highway with several cars visible ahead, illuminated mainly by taillights and headlights.",
+      "sub_caption": "dark SUV: A dark SUV traveling in the left lane, with visible red taillights reflecting the twilight.. Scene role: driving in the adjacent lane",
+      "measured_bbox": [
+        0.2594,
+        0.4853,
+        0.417,
+        0.6419
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_traveling_dark_suv.png",
+      "raw_ref_image": "references/raw_ref_traveling_dark_suv_attempt_01.png",
+      "reference_verify": "references/reference_verify_traveling_dark_suv.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_traveling_dark_suv_attempt_01.png",
+        "output": "references/ref_traveling_dark_suv.png",
+        "mask": "references/sam_mask_traveling_dark_suv.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          119.0,
+          198.0,
+          910.0,
+          810.0
+        ],
+        "mask_score": 3.470329,
+        "mask_area_ratio": 0.300606,
+        "elapsed_seconds": 8.5072
+      }
+    },
+    {
+      "name": "street_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c84f848e-2a5e0737:object:2",
+      "source_name": "street light",
+      "source_description": "Tall pole with a bright light on top, illuminating the road from the right side. Source dataset: BDD100K. Scene context: Nighttime driving view on a multi-lane highway with streetlights illuminating the dark road ahead.",
+      "sub_caption": "street light: Tall pole with a bright light on top, illuminating the road from the right side.. Scene role: providing illumination from the right sidewalk",
+      "measured_bbox": [
+        0.5577,
+        0.0219,
+        0.6964,
+        0.588
+      ],
+      "detection_confidence": 1.0,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_light.png",
+      "raw_ref_image": "references/raw_ref_street_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_light_attempt_01.png",
+        "output": "references/ref_street_light.png",
+        "mask": "references/sam_mask_street_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          330.0,
+          17.0,
+          688.0,
+          996.0
+        ],
+        "mask_score": 3.395182,
+        "mask_area_ratio": 0.033435,
+        "elapsed_seconds": 7.0701
+      }
+    },
+    {
+      "name": "vehicle_dashboard",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b5032e1d-dad95b60:object:9",
+      "source_name": "dashboard",
+      "source_description": "The dark, reflective dashboard and lower windshield area of the camera vehicle in the foreground. Source dataset: BDD100K. Scene context: View from inside a vehicle driving down a residential city street lined with trees, parked cars, and multi-story brick buildings.",
+      "sub_caption": "dashboard: The dark, reflective dashboard and lower windshield area of the camera vehicle in the foreground.. Scene role: anchoring the bottom of the frame to establish a driver's perspective",
+      "measured_bbox": [
+        0.0,
+        0.8881,
+        1.0,
+        1.0
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_vehicle_dashboard.png",
+      "raw_ref_image": "references/raw_ref_vehicle_dashboard_attempt_01.png",
+      "reference_verify": "references/reference_verify_vehicle_dashboard.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_vehicle_dashboard_attempt_01.png",
+        "output": "references/ref_vehicle_dashboard.png",
+        "mask": "references/sam_mask_vehicle_dashboard.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          223.0,
+          1023.0,
+          700.0
+        ],
+        "mask_score": 2.938032,
+        "mask_area_ratio": 0.282133,
+        "elapsed_seconds": 7.1679
+      }
+    },
+    {
+      "name": "drainage_grate",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b4d0e72d-3b208072:object:16",
+      "source_name": "drainage grate",
+      "source_description": "A metal drainage grate on the edge of the road on the right. Source dataset: BDD100K. Scene context: A driving scene on a multi-lane highway with a dark red minivan in the left lane, under a partly cloudy sky.",
+      "sub_caption": "drainage grate: A metal drainage grate on the edge of the road on the right.. Scene role: embedded in the road surface near the right curb",
+      "measured_bbox": [
+        0.5682,
+        0.6773,
+        0.8089,
+        0.73
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_drainage_grate.png",
+      "raw_ref_image": "references/raw_ref_drainage_grate_attempt_01.png",
+      "reference_verify": "references/reference_verify_drainage_grate.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_drainage_grate_attempt_01.png",
+        "output": "references/ref_drainage_grate.png",
+        "mask": "references/sam_mask_drainage_grate.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          250.0,
+          1023.0,
+          773.0
+        ],
+        "mask_score": 3.366042,
+        "mask_area_ratio": 0.379179,
+        "elapsed_seconds": 8.3171
+      }
+    },
+    {
+      "name": "white_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b3a102ed-6ef54f5e:object:3",
+      "source_name": "white car",
+      "source_description": "A white car visible further down the road in the right lane. Source dataset: BDD100K. Scene context: Nighttime driving scene in a city with cars stopped in traffic and a construction site on the right.",
+      "sub_caption": "white car: A white car visible further down the road in the right lane.. Scene role: driving ahead in the same lane",
+      "measured_bbox": [
+        0.4356,
+        0.5036,
+        0.4784,
+        0.548
+      ],
+      "detection_confidence": "high",
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_white_car.png",
+      "raw_ref_image": "references/raw_ref_white_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_white_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_white_car_attempt_01.png",
+        "output": "references/ref_white_car.png",
+        "mask": "references/sam_mask_white_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          137.0,
+          215.0,
+          884.0,
+          819.0
+        ],
+        "mask_score": 3.442096,
+        "mask_area_ratio": 0.295652,
+        "elapsed_seconds": 7.1564
+      }
+    },
+    {
+      "name": "yellow_lines",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c417a291-7802692d:object:8",
+      "source_name": "yellow lines",
+      "source_description": "Double yellow painted lines separating opposite directions of traffic. Source dataset: BDD100K. Scene context: A pedestrian crossing a street at a crosswalk with several cars stopped around them, with storefronts visible in the background.",
+      "sub_caption": "yellow lines: Double yellow painted lines separating opposite directions of traffic.. Scene role: painted down the center of the road",
+      "measured_bbox": [
+        0.0,
+        0.622,
+        0.2642,
+        0.7692
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_yellow_lines.png",
+      "raw_ref_image": "references/raw_ref_yellow_lines_attempt_01.png",
+      "reference_verify": "references/reference_verify_yellow_lines.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_yellow_lines_attempt_01.png",
+        "output": "references/ref_yellow_lines.png",
+        "mask": "references/sam_mask_yellow_lines.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          0.0,
+          1023.0,
+          1023.0
+        ],
+        "mask_score": 3.166027,
+        "mask_area_ratio": 0.242679,
+        "elapsed_seconds": 7.0941
+      }
+    },
+    {
+      "name": "street_trees",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c4891df0-24371ae1:object:3",
+      "source_name": "trees",
+      "source_description": "Numerous trees with dense green and yellowish foliage lining both sides of the road. Source dataset: BDD100K. Scene context: A dark SUV drives ahead on a wet road lined with trees and a fence under a cloudy sky.",
+      "sub_caption": "trees: Numerous trees with dense green foliage lining both sides of the road.. Scene role: growing along the sidewalks, adding greenery",
+      "measured_bbox": [
+        0.2664,
+        0.0,
+        0.7141,
+        0.5127
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_trees.png",
+      "raw_ref_image": "references/raw_ref_street_trees_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_trees.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_trees_attempt_01.png",
+        "output": "references/ref_street_trees.png",
+        "mask": "references/sam_mask_street_trees.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          35.0,
+          55.0,
+          1002.0,
+          1000.0
+        ],
+        "mask_score": 3.226043,
+        "mask_area_ratio": 0.439437,
+        "elapsed_seconds": 7.0986
+      }
+    },
+    {
+      "name": "twilight_sky",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c13c0d1f-00dfd075:object:9",
+      "source_name": "sky",
+      "source_description": "Clear twilight sky transitioning from bright near the horizon to dark blue at the top. Source dataset: BDD100K. Scene context: View from a moving vehicle looking down a multi-lane city street lined with parked cars and large trees under a clear twilight sky.",
+      "sub_caption": "sky: Clear twilight sky transitioning from bright near the horizon to dark blue at the top.. Scene role: visible above the buildings and trees at the end of the road",
+      "measured_bbox": [
+        0.188,
+        0.0,
+        0.862,
+        0.4846
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_twilight_sky.png",
+      "raw_ref_image": "references/raw_ref_twilight_sky_attempt_01.png",
+      "reference_verify": "references/reference_verify_twilight_sky.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_twilight_sky_attempt_01.png",
+        "output": "references/ref_twilight_sky.png",
+        "mask": "references/sam_mask_twilight_sky.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          72.0,
+          72.0,
+          951.0,
+          951.0
+        ],
+        "mask_score": 3.471577,
+        "mask_area_ratio": 0.631801,
+        "elapsed_seconds": 7.5016
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000003.json

@@ -0,0 +1,164 @@
+{
+  "sample_id": "sample_000003",
+  "target_total": 3,
+  "target_people": 1,
+  "target_objects": 2,
+  "canvas_size": [
+    1280,
+    720
+  ],
+  "canvas_aspect_ratio": "16:9",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 3,
+  "n_detected": 3,
+  "n_subjects": 3,
+  "subjects": [
+    {
+      "name": "waiting_pedestrian",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_12/282555,5c403000efcca35d.jpg:person:13",
+      "source_name": "shopper",
+      "source_description": "A person standing on the top level, wearing a dark top and dark pants. Source dataset: CrowdHuman. Scene context: A multi-level outdoor shopping mall with various people walking and shopping.",
+      "sub_caption": "shopper: A pedestrian wearing a dark top and dark pants, standing upright with a natural posture.. Scene role: Waiting at the curb near the crosswalk on the left side of the street.",
+      "measured_bbox": [
+        0.0928,
+        0.1174,
+        0.205,
+        0.9401
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_waiting_pedestrian.png",
+      "raw_ref_image": "references/raw_ref_waiting_pedestrian_attempt_01.png",
+      "reference_verify": "references/reference_verify_waiting_pedestrian.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_waiting_pedestrian_attempt_01.png",
+        "output": "references/ref_waiting_pedestrian.png",
+        "mask": "references/sam_mask_waiting_pedestrian.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          348.0,
+          48.0,
+          704.0,
+          1015.0
+        ],
+        "mask_score": 3.427649,
+        "mask_area_ratio": 0.155239,
+        "elapsed_seconds": 6.9951
+      }
+    },
+    {
+      "name": "black_sedan",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c0c183ff-1b24f541:object:7",
+      "source_name": "black sedan",
+      "source_description": "A black sedan driving in the right lane ahead of the red hatchback. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a multi-lane city street flanked by tall buildings, with various cars moving in the same direction and parked along the side.",
+      "sub_caption": "black sedan: A glossy black sedan with visible headlights and a detailed front grille.. Scene role: Approaching the crosswalk in the center traffic lane.",
+      "measured_bbox": [
+        0.3895,
+        0.2431,
+        0.591,
+        0.5084
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_black_sedan.png",
+      "raw_ref_image": "references/raw_ref_black_sedan_attempt_01.png",
+      "reference_verify": "references/reference_verify_black_sedan.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_black_sedan_attempt_01.png",
+        "output": "references/ref_black_sedan.png",
+        "mask": "references/sam_mask_black_sedan.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          221.0,
+          1023.0,
+          796.0
+        ],
+        "mask_score": 3.446312,
+        "mask_area_ratio": 0.340465,
+        "elapsed_seconds": 7.2258
+      }
+    },
+    {
+      "name": "silver_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:be49ae7a-1ffaa683:object:2",
+      "source_name": "silver car",
+      "source_description": "A silver car driving in the adjacent right lane, slightly ahead of the white car. Source dataset: BDD100K. Scene context: View from inside a vehicle driving in heavy traffic on a multi-lane road under a clear sky.",
+      "sub_caption": "silver car: A metallic silver car reflecting daylight.. Scene role: Driving in the right lane slightly ahead of the black sedan, approaching the intersection.",
+      "measured_bbox": [
+        0.6628,
+        0.2419,
+        0.9089,
+        0.4999
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_silver_car.png",
+      "raw_ref_image": "references/raw_ref_silver_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_silver_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_silver_car_attempt_01.png",
+        "output": "references/ref_silver_car.png",
+        "mask": "references/sam_mask_silver_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          13.0,
+          220.0,
+          1011.0,
+          811.0
+        ],
+        "mask_score": 3.077144,
+        "mask_area_ratio": 0.338042,
+        "elapsed_seconds": 7.0902
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000004.json

@@ -0,0 +1,256 @@
+{
+  "sample_id": "sample_000004",
+  "target_total": 5,
+  "target_people": 1,
+  "target_objects": 4,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 5,
+  "n_detected": 5,
+  "n_subjects": 5,
+  "subjects": [
+    {
+      "name": "walker",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_51/273275,145927000354f7525.jpg:person:10",
+      "source_name": "walker",
+      "source_description": "A man walking, wearing a blue hoodie with 'EMO' written on it, grey pants, and dark shoes. Source dataset: CrowdHuman. Scene context: A group of people, some wearing matching white t-shirts, are walking and jogging along a paved path next to a road, with a grey SUV parked on a grassy hill in the background.",
+      "sub_caption": "walker: A man walking, wearing a plain blue hoodie, grey pants, and dark shoes.. Scene role: Standing at the edge of the sidewalk near the crosswalk, waiting to cross the street.",
+      "measured_bbox": [
+        0.7914,
+        0.2893,
+        0.834,
+        0.4815
+      ],
+      "detection_confidence": 1.0,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_walker.png",
+      "raw_ref_image": "references/raw_ref_walker_attempt_01.png",
+      "reference_verify": "references/reference_verify_walker.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_walker_attempt_01.png",
+        "output": "references/ref_walker.png",
+        "mask": "references/sam_mask_walker.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          336.0,
+          51.0,
+          688.0,
+          1005.0
+        ],
+        "mask_score": 3.433924,
+        "mask_area_ratio": 0.16005,
+        "elapsed_seconds": 7.2846
+      }
+    },
+    {
+      "name": "traffic_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b6df605f-51c158b8:object:6",
+      "source_name": "traffic light",
+      "source_description": "A set of traffic lights suspended over the intersection, showing a red light. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a city street on a sunny day with a U-Haul truck in the opposite lane, parked cars along the right curb, and pedestrians crossing a crosswalk.",
+      "sub_caption": "traffic light: A set of traffic lights suspended over the intersection, showing a red light.. Scene role: Hanging high above the intersection in the upper-center of the frame.",
+      "measured_bbox": [
+        0.4425,
+        0.023,
+        0.467,
+        0.1052
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_traffic_light.png",
+      "raw_ref_image": "references/raw_ref_traffic_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_traffic_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_traffic_light_attempt_01.png",
+        "output": "references/ref_traffic_light.png",
+        "mask": "references/sam_mask_traffic_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          160.0,
+          93.0,
+          864.0,
+          930.0
+        ],
+        "mask_score": 3.437579,
+        "mask_area_ratio": 0.253583,
+        "elapsed_seconds": 7.0663
+      }
+    },
+    {
+      "name": "delivery_truck",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b3a7b21a-48bcf2b8:object:2",
+      "source_name": "delivery truck",
+      "source_description": "A large white box truck parked behind the fence on the right, with visible green and purple logos. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a wide, paved road flanked by an industrial area with fences and delivery trucks under a cloudy sky.",
+      "sub_caption": "delivery truck: A large, plain white box delivery truck.. Scene role: Parked alongside the right curb in the background, past the intersection.",
+      "measured_bbox": [
+        0.576,
+        0.1929,
+        0.7135,
+        0.4081
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_delivery_truck.png",
+      "raw_ref_image": "references/raw_ref_delivery_truck_attempt_01.png",
+      "reference_verify": "references/reference_verify_delivery_truck.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_delivery_truck_attempt_01.png",
+        "output": "references/ref_delivery_truck.png",
+        "mask": "references/sam_mask_delivery_truck.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          95.0,
+          100.0,
+          910.0,
+          932.0
+        ],
+        "mask_score": 3.445823,
+        "mask_area_ratio": 0.476913,
+        "elapsed_seconds": 7.1923
+      }
+    },
+    {
+      "name": "dark_parked_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bcd37eef-1b958ae3:object:5",
+      "source_name": "dark parked car",
+      "source_description": "Another dark-colored car parked on the right side of the street. Source dataset: BDD100K. Scene context: A nighttime view from a vehicle driving down a multi-lane city street with parked cars on the right and oncoming traffic on the left.",
+      "sub_caption": "dark parked car: A dark-colored sedan.. Scene role: Parked on the right side of the street near the sidewalk in the mid-ground.",
+      "measured_bbox": [
+        0.8414,
+        0.3717,
+        0.9967,
+        0.7454
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_dark_parked_car.png",
+      "raw_ref_image": "references/raw_ref_dark_parked_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_dark_parked_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_dark_parked_car_attempt_01.png",
+        "output": "references/ref_dark_parked_car.png",
+        "mask": "references/sam_mask_dark_parked_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          19.0,
+          336.0,
+          1003.0,
+          700.0
+        ],
+        "mask_score": 3.408233,
+        "mask_area_ratio": 0.181406,
+        "elapsed_seconds": 8.4178
+      }
+    },
+    {
+      "name": "street_lines",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c0c9ec9a-d3638a82:object:6",
+      "source_name": "street lines",
+      "source_description": "Double yellow lines separating traffic directions and white painted lines indicating lanes and crosswalks. Source dataset: BDD100K. Scene context: A view from inside a car driving on a city street, approaching an intersection underneath an elevated railway structure.",
+      "sub_caption": "street lines: Double yellow lines separating traffic directions and crisp white painted lines indicating lanes and a crosswalk.. Scene role: Painted on the asphalt road surface, extending from the foreground toward the intersection.",
+      "measured_bbox": [
+        0.003,
+        0.3541,
+        0.915,
+        0.8612
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_lines.png",
+      "raw_ref_image": "references/raw_ref_street_lines_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_lines.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_street_lines_attempt_01.png",
+        "output": "references/ref_street_lines.png",
+        "mask": "references/sam_mask_street_lines.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          38.0,
+          225.0,
+          985.0,
+          799.0
+        ],
+        "mask_score": 3.287982,
+        "mask_area_ratio": 0.400985,
+        "elapsed_seconds": 7.2613
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000005.json

@@ -0,0 +1,302 @@
+{
+  "sample_id": "sample_000005",
+  "target_total": 6,
+  "target_people": 3,
+  "target_objects": 3,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 6,
+  "n_detected": 6,
+  "n_subjects": 6,
+  "subjects": [
+    {
+      "name": "firefighter",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_2/282555,e0af90003451118a.jpg:person:8",
+      "source_name": "firefighter",
+      "source_description": "Wearing a dark uniform with yellow reflective stripes and a white helmet, standing facing away near the fire truck. Source dataset: CrowdHuman. Scene context: Emergency response personnel, including firefighters and ambulance crew, are gathered outside a large classical building with pillars and banners, accompanied by emergency vehicles.",
+      "sub_caption": "firefighter: Wearing a dark uniform with yellow reflective stripes and a white helmet, standing facing away.. Scene role: Assisting with incident management, positioned near the stopped car and barrier.",
+      "measured_bbox": [
+        0.2626,
+        0.3463,
+        0.3289,
+        0.6561
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_firefighter.png",
+      "raw_ref_image": "references/raw_ref_firefighter_attempt_01.png",
+      "reference_verify": "references/reference_verify_firefighter.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_firefighter_attempt_01.png",
+        "output": "references/ref_firefighter.png",
+        "mask": "references/sam_mask_firefighter.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          317.0,
+          34.0,
+          709.0,
+          1009.0
+        ],
+        "mask_score": 3.445343,
+        "mask_area_ratio": 0.178691,
+        "elapsed_seconds": 7.0362
+      }
+    },
+    {
+      "name": "uniformed_officer",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_1/273275,f68c20007e0bf148.jpg:person:3",
+      "source_name": "uniformed officer",
+      "source_description": "wearing a khaki uniform and helmet, holding a baton, looking towards the left Source dataset: CrowdHuman. Scene context: A large crowd of people, including some in uniform with batons and helmets, stands in front of a red and yellow building.",
+      "sub_caption": "uniformed officer: Wearing a khaki uniform and helmet, holding a baton, looking towards the left.. Scene role: Directing surrounding traffic away from the stopped vehicle using a baton.",
+      "measured_bbox": [
+        0.0497,
+        0.3566,
+        0.1691,
+        0.6118
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_uniformed_officer.png",
+      "raw_ref_image": "references/raw_ref_uniformed_officer_attempt_01.png",
+      "reference_verify": "references/reference_verify_uniformed_officer.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_uniformed_officer_attempt_01.png",
+        "output": "references/ref_uniformed_officer.png",
+        "mask": "references/sam_mask_uniformed_officer.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          331.0,
+          24.0,
+          689.0,
+          1005.0
+        ],
+        "mask_score": 3.475629,
+        "mask_area_ratio": 0.156165,
+        "elapsed_seconds": 7.0984
+      }
+    },
+    {
+      "name": "bystander_in_suit",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_28/273278,b62280001bda6f1a.jpg:person:19",
+      "source_name": "crowd member",
+      "source_description": "A person far right in the background wearing a suit. Source dataset: CrowdHuman. Scene context: A crowd of people gathers in front of an old stone building with a prominent arched doorway and two large animal statues.",
+      "sub_caption": "crowd member: A person wearing a suit.. Scene role: Standing off to the right side of the road behind the barrier, acting as the driver or an involved pedestrian.",
+      "measured_bbox": [
+        0.7467,
+        0.3318,
+        0.8036,
+        0.5111
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_bystander_in_suit.png",
+      "raw_ref_image": "references/raw_ref_bystander_in_suit_attempt_01.png",
+      "reference_verify": "references/reference_verify_bystander_in_suit.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_bystander_in_suit_attempt_01.png",
+        "output": "references/ref_bystander_in_suit.png",
+        "mask": "references/sam_mask_bystander_in_suit.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          341.0,
+          59.0,
+          677.0,
+          996.0
+        ],
+        "mask_score": 3.480669,
+        "mask_area_ratio": 0.144797,
+        "elapsed_seconds": 7.0242
+      }
+    },
+    {
+      "name": "traffic_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_36/273275,6a11d000f52c34a9.jpg:object:0",
+      "source_name": "traffic light",
+      "source_description": "A black multi-lens traffic light fixture mounted on a pole above the street. Source dataset: CrowdHuman. Scene context: A male tour guide is speaking to a group of people standing on a city sidewalk next to a road crossing.",
+      "sub_caption": "traffic light: A black multi-lens traffic light fixture mounted on a pole above the street.. Scene role: Hanging overhead or mounted prominently on a pole at the intersection.",
+      "measured_bbox": [
+        0.5381,
+        0.0316,
+        0.5856,
+        0.2076
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_traffic_light.png",
+      "raw_ref_image": "references/raw_ref_traffic_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_traffic_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_traffic_light_attempt_01.png",
+        "output": "references/ref_traffic_light.png",
+        "mask": "references/sam_mask_traffic_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          272.0,
+          15.0,
+          750.0,
+          1006.0
+        ],
+        "mask_score": 3.448339,
+        "mask_area_ratio": 0.303974,
+        "elapsed_seconds": 8.3734
+      }
+    },
+    {
+      "name": "concrete_barrier",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c946c532-07177e0a:object:11",
+      "source_name": "concrete barrier",
+      "source_description": "A continuous low concrete wall acting as a barrier on the right side of the road. Source dataset: BDD100K. Scene context: View from inside a vehicle driving on a multi-lane highway during the day, with construction or industrial sites visible alongside.",
+      "sub_caption": "concrete barrier: A continuous low concrete wall acting as a barrier on the right side of the road.. Scene role: Lining the right side of the street, separating the pedestrian walkway or construction zone from the active traffic lane.",
+      "measured_bbox": [
+        0.6322,
+        0.4972,
+        0.9964,
+        0.6985
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_concrete_barrier.png",
+      "raw_ref_image": "references/raw_ref_concrete_barrier_attempt_01.png",
+      "reference_verify": "references/reference_verify_concrete_barrier.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_concrete_barrier_attempt_01.png",
+        "output": "references/ref_concrete_barrier.png",
+        "mask": "references/sam_mask_concrete_barrier.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          53.0,
+          219.0,
+          970.0,
+          811.0
+        ],
+        "mask_score": 3.469119,
+        "mask_area_ratio": 0.3653,
+        "elapsed_seconds": 7.0274
+      }
+    },
+    {
+      "name": "silver_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:be3d3a81-326a032d:object:0",
+      "source_name": "silver car",
+      "source_description": "A silver compact hatchback car facing forward, waiting at an intersection. Its brake lights are on. Source dataset: BDD100K. Scene context: A rainy street scene showing cars waiting at an intersection surrounded by tall buildings and urban infrastructure.",
+      "sub_caption": "silver car: A silver compact hatchback car facing forward, waiting at an intersection with illuminated brake lights.. Scene role: Stopped in the active lane near the barrier, serving as the focal point of the traffic response.",
+      "measured_bbox": [
+        0.3396,
+        0.3754,
+        0.6399,
+        0.6647
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_silver_car.png",
+      "raw_ref_image": "references/raw_ref_silver_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_silver_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_silver_car_attempt_01.png",
+        "output": "references/ref_silver_car.png",
+        "mask": "references/sam_mask_silver_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          46.0,
+          215.0,
+          976.0,
+          829.0
+        ],
+        "mask_score": 3.457698,
+        "mask_area_ratio": 0.330622,
+        "elapsed_seconds": 7.0933
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CityPersons_CrowdHuman_samples/rows/sample_000006.json

@@ -0,0 +1,394 @@
+{
+  "sample_id": "sample_000006",
+  "target_total": 8,
+  "target_people": 1,
+  "target_objects": 7,
+  "canvas_size": [
+    1280,
+    720
+  ],
+  "canvas_aspect_ratio": "16:9",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 8,
+  "n_detected": 8,
+  "n_subjects": 8,
+  "subjects": [
+    {
+      "name": "distant_pedestrian",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_73/283991,17cd800008079067.jpg:person:18",
+      "source_name": "pedestrian",
+      "source_description": "Another person in the distant background near the green structure. Source dataset: CrowdHuman. Scene context: A large crowd of people walking across a street with trees, streetlamps, and classic architecture in the background.",
+      "sub_caption": "pedestrian: A person walking across the street in the distant background.. Scene role: Crossing the crosswalk in the distance ahead of the approaching vehicles.",
+      "measured_bbox": [
+        0.3877,
+        0.478,
+        0.4204,
+        0.5881
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_distant_pedestrian.png",
+      "raw_ref_image": "references/raw_ref_distant_pedestrian_attempt_01.png",
+      "reference_verify": "references/reference_verify_distant_pedestrian.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_distant_pedestrian_attempt_01.png",
+        "output": "references/ref_distant_pedestrian.png",
+        "mask": "references/sam_mask_distant_pedestrian.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          324.0,
+          9.0,
+          705.0,
+          1015.0
+        ],
+        "mask_score": 3.338419,
+        "mask_area_ratio": 0.174056,
+        "elapsed_seconds": 8.694
+      }
+    },
+    {
+      "name": "vertical_illuminated_sign",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bb1b7e42-9608265e:object:6",
+      "source_name": "street sign",
+      "source_description": "A vertical 'PARK' sign illuminated on the right side of the street, indicating a parking garage. Source dataset: BDD100K. Scene context: A dashcam view from a vehicle driving down a city street with tall buildings on both sides, following a yellow taxi, with other cars parked and driving.",
+      "sub_caption": "street sign: A vertical illuminated neon sign with abstract shapes, glowing brightly.. Scene role: Mounted on the building facade on the right side of the street, adding ambient night lighting.",
+      "measured_bbox": [
+        0.7683,
+        0.0355,
+        0.8177,
+        0.2837
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_vertical_illuminated_sign.png",
+      "raw_ref_image": "references/raw_ref_vertical_illuminated_sign_attempt_01.png",
+      "reference_verify": "references/reference_verify_vertical_illuminated_sign.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_vertical_illuminated_sign_attempt_01.png",
+        "output": "references/ref_vertical_illuminated_sign.png",
+        "mask": "references/sam_mask_vertical_illuminated_sign.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          305.0,
+          20.0,
+          728.0,
+          1002.0
+        ],
+        "mask_score": 3.37343,
+        "mask_area_ratio": 0.273593,
+        "elapsed_seconds": 7.1332
+      }
+    },
+    {
+      "name": "emergency_vehicle",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b99f250d-886111c5:object:5",
+      "source_name": "vehicle",
+      "source_description": "A dark-colored vehicle partially visible in the left background with blue emergency lights flashing. Source dataset: BDD100K. Scene context: A nighttime city street intersection showing a crosswalk, construction barriers, and illuminated traffic signals.",
+      "sub_caption": "vehicle: A dark-colored vehicle with bright blue emergency lights flashing.. Scene role: Parked on the left side of the street near the intersection.",
+      "measured_bbox": [
+        0.1031,
+        0.4564,
+        0.2827,
+        0.6497
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_emergency_vehicle.png",
+      "raw_ref_image": "references/raw_ref_emergency_vehicle_attempt_01.png",
+      "reference_verify": "references/reference_verify_emergency_vehicle.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_emergency_vehicle_attempt_01.png",
+        "output": "references/ref_emergency_vehicle.png",
+        "mask": "references/sam_mask_emergency_vehicle.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          8.0,
+          237.0,
+          1015.0,
+          828.0
+        ],
+        "mask_score": 3.468468,
+        "mask_area_ratio": 0.355034,
+        "elapsed_seconds": 7.0896
+      }
+    },
+    {
+      "name": "white_panel_van",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b5047c50-e1facff6:object:2",
+      "source_name": "white van",
+      "source_description": "A large white panel van with red taillights illuminated, driving in the right lane ahead. Source dataset: BDD100K. Scene context: View from inside a vehicle driving down a multi-lane city street on a sunny day with moderate traffic.",
+      "sub_caption": "white van: A large white panel van with red taillights illuminated.. Scene role: Driving in the lane directly ahead of the camera perspective.",
+      "measured_bbox": [
+        0.4556,
+        0.3288,
+        0.5926,
+        0.6597
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_white_panel_van.png",
+      "raw_ref_image": "references/raw_ref_white_panel_van_attempt_01.png",
+      "reference_verify": "references/reference_verify_white_panel_van.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_white_panel_van_attempt_01.png",
+        "output": "references/ref_white_panel_van.png",
+        "mask": "references/sam_mask_white_panel_van.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          181.0,
+          63.0,
+          843.0,
+          937.0
+        ],
+        "mask_score": 2.636854,
+        "mask_area_ratio": 0.376409,
+        "elapsed_seconds": 7.1379
+      }
+    },
+    {
+      "name": "double_solid_line",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bc886d37-5b22c313:object:7",
+      "source_name": "double solid white line",
+      "source_description": "Two continuous white painted lines on the dark asphalt road surface, separating the two lanes of traffic. Source dataset: BDD100K. Scene context: A view from inside a car driving through a brightly lit tunnel with tiled walls, following a silver SUV and a dark compact car.",
+      "sub_caption": "double solid white line: Two continuous white painted lines on the dark asphalt road surface.. Scene role: Separating the traffic lanes on the dark road, leading toward the intersection.",
+      "measured_bbox": [
+        0.1922,
+        0.6133,
+        0.4541,
+        1.0
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_double_solid_line.png",
+      "raw_ref_image": "references/raw_ref_double_solid_line_attempt_01.png",
+      "reference_verify": "references/reference_verify_double_solid_line.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_double_solid_line_attempt_01.png",
+        "output": "references/ref_double_solid_line.png",
+        "mask": "references/sam_mask_double_solid_line.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          22.0,
+          186.0,
+          1001.0,
+          837.0
+        ],
+        "mask_score": 3.460181,
+        "mask_area_ratio": 0.372935,
+        "elapsed_seconds": 8.3174
+      }
+    },
+    {
+      "name": "dark_building_facade",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c807cb19-7e09cb11:object:8",
+      "source_name": "building facade",
+      "source_description": "Dark outlines of buildings lining the street on both sides, with some lit windows. Source dataset: BDD100K. Scene context: Nighttime driving view on a multi-lane city street with traffic lights and vehicles ahead.",
+      "sub_caption": "building facade: Dark outlines of buildings with scattered, warm-toned lit windows.. Scene role: Forming the urban backdrop along the left side of the street.",
+      "measured_bbox": [
+        0.1397,
+        0.0,
+        0.366,
+        0.5427
+      ],
+      "detection_confidence": 0.8,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_dark_building_facade.png",
+      "raw_ref_image": "references/raw_ref_dark_building_facade_attempt_01.png",
+      "reference_verify": "references/reference_verify_dark_building_facade.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_dark_building_facade_attempt_01.png",
+        "output": "references/ref_dark_building_facade.png",
+        "mask": "references/sam_mask_dark_building_facade.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          128.0,
+          0.0,
+          887.0,
+          1000.0
+        ],
+        "mask_score": 2.829968,
+        "mask_area_ratio": 0.624767,
+        "elapsed_seconds": 7.1675
+      }
+    },
+    {
+      "name": "awning_building_corner",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c06d23aa-cb9ae751:object:6",
+      "source_name": "building corner",
+      "source_description": "The corner of a building on the right side, with an orange or red awning and some lit signs. Source dataset: BDD100K. Scene context: Nighttime driving scene at an intersection with a stop sign and a large black SUV passing on the right.",
+      "sub_caption": "building corner: The corner of a building featuring an awning and brightly lit abstract signboards.. Scene role: Anchoring the right side of the intersection with a warm architectural glow.",
+      "measured_bbox": [
+        0.6102,
+        0.3347,
+        0.7867,
+        0.5412
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_awning_building_corner.png",
+      "raw_ref_image": "references/raw_ref_awning_building_corner_attempt_01.png",
+      "reference_verify": "references/reference_verify_awning_building_corner.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_awning_building_corner_attempt_01.png",
+        "output": "references/ref_awning_building_corner.png",
+        "mask": "references/sam_mask_awning_building_corner.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          27.0,
+          27.0,
+          975.0,
+          980.0
+        ],
+        "mask_score": 3.458235,
+        "mask_area_ratio": 0.594922,
+        "elapsed_seconds": 7.3072
+      }
+    },
+    {
+      "name": "green_street_sign",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_50/273278,febe100057ca94db.jpg:object:5",
+      "source_name": "street sign",
+      "source_description": "A green street sign with white text visible on the left side. Source dataset: CrowdHuman. Scene context: Several pedestrians are crossing a street at a crosswalk on a sunny day.",
+      "sub_caption": "street sign: A standard green street sign without any readable text.. Scene role: Hanging from a traffic light pole near the intersection.",
+      "measured_bbox": [
+        0.5754,
+        0.1583,
+        0.6522,
+        0.1884
+      ],
+      "detection_confidence": 100,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_green_street_sign.png",
+      "raw_ref_image": "references/raw_ref_green_street_sign_attempt_01.png",
+      "reference_verify": "references/reference_verify_green_street_sign.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/driving/samples/sample_000006/references/raw_ref_green_street_sign_attempt_01.png",
+        "output": "references/ref_green_street_sign.png",
+        "mask": "references/sam_mask_green_street_sign.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/dataset/tmp/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          61.0,
+          378.0,
+          962.0,
+          645.0
+        ],
+        "mask_score": 3.379525,
+        "mask_area_ratio": 0.536634,
+        "elapsed_seconds": 7.1734
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}