Researchers have introduced SurgAtlas, a comprehensive dataset for surgical video-language understanding, featuring over 2,391 hours of open and minimally invasive surgery footage. This dataset is the largest of its kind and the first to extensively cover open surgery procedures. SurgAtlas includes diverse annotations, such as segment-level captions and question-answer pairs, generated through an automated pipeline enhanced by LLMs. The dataset has been used to fine-tune the Qwen3-VL-8B model, achieving competitive results on established surgical benchmarks and paving the way for advanced surgical AI systems. AI
IMPACT Enables training of advanced surgical foundation models and next-generation multimodal AI systems for surgery.
RANK_REASON The cluster describes a new dataset and its use in fine-tuning a model, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
