CoolFace
Datasetpublic

nazzzz5265/rav4-video-detection

RAV4 Video Detections Object detection index and image-to-video retrieval results for the Toyota RAV4 exterior video (YcvECxtXoxQ). Files File Description video_detections.parquet All car-part detections from the video corpus (92 rows) retrieval_results.parquet Retrieval results matching query images to video segments (72 rows) Schema: video_detections.parquet Column Type Description video_id string YouTube video ID… See the full description on the dataset page: https://huggingface.co/datasets/nazzzz5265/rav4-video-detection.

sourceHugging Facemitupdated 7mo agoView on Hugging Face
0likes4downloads
Dataset Card

RAV4 Video Detections

Object detection index and image-to-video retrieval results for the Toyota RAV4 exterior video (YcvECxtXoxQ).

Files

FileDescription
video_detections.parquetAll car-part detections from the video corpus (92 rows)
retrieval_results.parquetRetrieval results matching query images to video segments (72 rows)

Schema: video_detections.parquet

ColumnTypeDescription
video_idstringYouTube video ID (YcvECxtXoxQ)
frame_idxintSampled frame number
timestamp_secintTime position in seconds (120–164, clip from 2:00–2:45)
class_labelstringDetected car part class (e.g. wheel, hood, front_bumper)
confidence_scorefloatDetector confidence (0–1)
x_minfloatBounding box left edge (pixels)
y_minfloatBounding box top edge (pixels)
x_maxfloatBounding box right edge (pixels)
y_maxfloatBounding box bottom edge (pixels)

Schema: retrieval_results.parquet

ColumnTypeDescription
query_indexintIndex of the query image in aegean-ai/rav4-exterior-images
query_timestampstringTimestamp of the query image (e.g. 00:05)
start_timestampintStart of matched video segment (seconds)
end_timestampintEnd of matched video segment (seconds)
class_labelstringCar part classes shared between query and video
number_of_supporting_detectionsintNumber of video frame detections supporting the match
verify_urlstringYouTube embed URL to verify the matched segment

Detector

  • —Model: YOLOv8n-seg (Ultralytics), fine-tuned on carparts-seg dataset
  • —Training: 50 epochs, 640px, Google Colab T4 GPU — mAP50: 69.3% (box), 71.0% (mask)
  • —Car part classes: backbumper, frontbumper, hood, wheel, frontdoor, backdoor, frontglass, backglass, frontlight, backlight, mirrors, trunk, tailgate, and others (23 total)
  • —Confidence threshold: 0.5
  • —Video sampling: 1 frame/second from a 45-second clip (2:00–2:45)

Query Dataset

Query images sourced from `aegean-ai/rav4-exterior-images`.