CoolFace
Datasetpublic

Bineet2407/rav4-video-detections

RAV4 Video Detections Car-part detections and retrieval results from the Toyota RAV4 exterior review video (YcvECxtXoxQ) using a two-pass smart sampling strategy: Coarse pass (1 frame / 5 s) scored by CLIP zero-shot exterior classifier (openai/clip-vit-base-patch32). Dense pass (1 fps) applied only to windows with exterior confidence >= 0.70. YOLOv8n-seg fine-tuned on the Ultralytics Car Parts Segmentation dataset runs on all dense frames. File 1:… See the full description on the dataset page: https://huggingface.co/datasets/Bineet2407/rav4-video-detections.

sourceHugging Facemitupdated 7mo agoView on Hugging Face
0likes11downloads
Dataset Card

RAV4 Video Detections

Car-part detections and retrieval results from the Toyota RAV4 exterior review video (YcvECxtXoxQ) using a two-pass smart sampling strategy:

  1. 1.Coarse pass (1 frame / 5 s) scored by CLIP zero-shot exterior classifier (openai/clip-vit-base-patch32).
  2. 2.Dense pass (1 fps) applied only to windows with exterior confidence >= 0.70.
  3. 3.YOLOv8n-seg fine-tuned on the Ultralytics Car Parts Segmentation dataset runs on all dense frames.

File 1: video_detections.parquet

One row per bounding-box detection in the video. This is the searchable index — 6,653 detections spanning 0 s – 2,791 s.

ColumnTypeDescription
video_idstrYouTube video ID (YcvECxtXoxQ)
frame_indexintTimestamp rounded to nearest second
timestamp_secfloatExact time position in seconds
class_idintDetector class index (0–22)
class_labelstrHuman-readable part name (e.g. wheel, hood)
x_minfloatBounding box left edge (pixels)
y_minfloatBounding box top edge (pixels)
x_maxfloatBounding box right edge (pixels)
y_maxfloatBounding box bottom edge (pixels)
confidence_scorefloatDetector confidence [0, 1]
detector_namestrModel identifier (yolov8n-seg-carparts)

File 2: retrieval_results.parquet

One row per retrieved video segment for a query image. 310 results across 65 query images from the aegean-ai/rav4-exterior-images dataset.

ColumnTypeDescription
query_indexintIndex of the query image in the dataset
query_timestampstrSource timestamp of the query image (MM:SS)
class_labelstrCar part class matched between query and video
start_timestampintSegment start time in seconds
end_timestampintSegment end time in seconds
number_of_supporting_detectionsintFrames within segment containing a detection of the matched class
youtube_linkstrYouTube embed URL with start/end parameters for direct verification