Surgenetyoutube: Surgical Video Frames Sampled from YouTube

A curated collection of surgical video frames derived from 680 hours of footage sourced from YouTube. The dataset was processed using code from Schmidgall et al. (2024) and sampled at 1 frame per second, with manual annotation applied. It was created by TimJaspersTue and last updated on January 17, —2025.

Use Cases

• Training surgical scene understanding models based on annotated video frames.
• Developing foundational models for medical AI based on a large-scale video corpus.
• Benchmarking video frame classification algorithms in the surgical domain.
• Conducting research on surgical workflow analysis using sampled video data.

Strengths

• Derived from 680 hours of source video footage.
• Manually annotated to remove non-minimally relevant content.
• Sampled at a consistent 1 frame per second rate.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset's license is unspecified, requiring verification before use.

Provenance

Source

YouTube

Collection Method

Video footage processed using code from Schmidgall et al. (2024), sampled at 1 fps, and manually annotated.

Time Range

null

Freshness

Last updated 2025-01-17 06:47:44; freshness should be verified.

Geography

null