swati6singh/ss20478_rav4-semantic-retrieval

Image-to-Video Semantic Retrieval (Assignment 2)

This dataset contains outputs for semantic retrieval of car exterior components in a Toyota RAV4 review video.

The system detects exterior components in video frames and retrieves relevant video segments for query images.

Files

1. detections.parquet

Contains structured detections for extracted video frames.

Schema:

• —frame_id (int) Sequential frame identifier.

• —detections (list[dict]) List of detections in that frame. Each detection contains:
• —video_id (str): YouTube video ID.
• —frame_index (int): frame index in the sampled clip.
• —timestamp_sec (int): timestamp in seconds in the original YouTube video.
• —class_label (str): detected car exterior component.
• —bounding_box (list[float]): [xmin, ymin, xmax, ymax].
• —confidence_score (float): detector confidence.

2. querytoclip.parquet

Final semantic retrieval results.

Schema:

• —query_id (int): query image index.
• —youtube_link (str): retrieved video clip in embed format.

Each query image is processed using the trained detector, matched against video detections, and mapped to a contiguous time segment.