Reproduce “Surgical Visual Understanding (SurgVU) Dataset” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Aneeq Zia, Max Berniker, Rogerio Nespolo, Xiaorui Zhang, Conor Perreault, Ziheng Wang, Benjamin Mueller, Ryan Schmidt and 3 more

The SurgVU dataset is introduced as a large-scale collection of over 840 hours of robotic surgical video from Intuitive Surgical, accompanied by diverse annotations including tool presence, surgical steps, detailed natural language descriptions, and visual-language Q&A pairs. This publicly available resource provides a new common benchmark for surgical data science, supporting advancements in real-time guidance, weakly supervised learning, activity recognition, performance quantification, and vision-language models.

Key insights

  • Robotic-assisted surgery generates extensive, high-resolution video data that can be leveraged for scientific advancement through machine learning.
  • Standardized surgical training environments offer a controlled setting for collecting consistent video data suitable for comprehensive dataset construction.
  • Integrating diverse annotation types, ranging from automatic system logs to detailed expert-curated textual descriptions, is crucial for enabling a broad spectrum of multimodal and vision-language machine learning research.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.