Datasetpublic

lin1318/object_detection

Image-to-Video Semantic Retrieval via Object Detection – Dataset Schema This repository contains detection and retrieval results for a Toyota RAV4 2026 review video. Frames were sampled every 5 seconds from the source video. 1. Video Detections File: video_detections.parquet Each row corresponds to a single object detection in a sampled video frame. Schema Column Type Description video_id string YouTube video identifier frame_index int… See the full description on the dataset page: https://huggingface.co/datasets/lin1318/object_detection.

sourceHugging Facemitupdated 7mo agoView on Hugging Face

0likes19downloads

Dataset Card

Image-to-Video Semantic Retrieval via Object Detection – Dataset Schema

This repository contains detection and retrieval results for a Toyota RAV4 2026 review video.

Frames were sampled every 5 seconds from the source video.

1. Video Detections

File: video_detections.parquet

Each row corresponds to a single object detection in a sampled video frame.

Schema

Column	Type	Description
video_id	string	YouTube video identifier
frame_index	int	Index of the sampled frame (starting from 1)
timestamp_sec	int	Time position in seconds
class_label	string	Detected exterior component label
bounding_box	list[float]	Bounding box in pixel coordinates: [xmin, ymin, xmax, ymax]
confidence_score	float	YOLO detection confidence score

2. Retrieval Results

File: query_retrieval_results.parquet

Each row corresponds to one retrieved video segment for a given query image and component.

Retrieval is performed by matching detected component labels between query image detections and the video detection index. Contiguous timestamps are merged into segments.

Schema

Column	Type	Description
query_index	int	Index of the query image
querytimestampsec	int	Timestamp of the query image in seconds
video_id	string	YouTube video identifier
class_label	string	Component label used for matching
queryclassconf	float	Detection confidence of the component in the query image
start_timestamp	int	Start time (seconds) of the retrieved segment
end_timestamp	int	End time (seconds) of the retrieved segment
numberofsupporting_detections	int	Number of consecutive frames supporting this segment
verify_url	string	YouTube embed URL for manual verification

Verification

Retrieved segments can be verified using:

https://www.youtube.com/embed/{videoid}?start={starttimestamp}&end={end_timestamp}