PotriAbhiB/rav4-video-detections

RAV4 Exterior Video Detections

This dataset contains object-part detections for the YouTube video corpus YcvECxtXoxQ (car exterior).

Files

• —video_detections.parquet: one row per detection.

Parquet schema

• —video_id (string): YouTube video id (YcvECxtXoxQ)
• —frame_index (int): extracted frame number (from frame_XXXXXX.jpg)
• —timestamp_sec (int): time in seconds in the source video
• —class_label (string): predicted exterior part class (e.g., wheel, front_bumper)
• —bounding_box (list[float]): [x_min, y_min, x_max, y_max] in pixel coordinates
• —confidence_score (float): model confidence score

Notes

Frames were sampled at 1 frame per 2 seconds. Detections were produced using a YOLO segmentation model fine-tuned on Ultralytics Car Parts Segmentation.

Source video

• —YouTube: https://www.youtube.com/watch?v=YcvECxtXoxQ

Sampling

• —1 frame every 2 seconds.
• —timestamp_sec = (frame_index - 1) * 2

Quick usage

import pandas as pd
df = pd.read_parquet("video_detections.parquet")
df[df["class_label"] == "wheel"].head()