Active Learning with Weak Supervision Cuts Surgical Video Annotation Effort by 50%, Penn Study Finds
Laparoscopes · x · 2026-11-11
A University of Pennsylvania engineering team published a short communication in International Journal of Computer Assisted Radiology and Surgery (IJCARS) proposing a human-in-the-loop knowledge acquisition framework that combines active learning with dual-loss optimization for localizing and segmenting tools in laparoscopic surgical videos.
Method: A foundation model generates temporally consistent class activation maps (CAMs) from video, trained with two complementary objectives — a weak supervision loss on video-level tool presence labels, and an image-level mask loss on human-corrected annotations from active learning. The pipeline iteratively proposes pseudo-masks that guide expert annotators to refine the model's knowledge, instead of requiring dense pixel-level annotation upfront.
Results: The framework reduces surgical video annotation effort by roughly 50% by the end of training versus fully manual annotation, making surgical tool segmentation datasets far more scalable to build.
Related event: UPenn Team Speeds Up Surgical Video Annotation by 50% with Active Learning(3 posts)→
More from Research
- CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques — tomssilver · 2026-11-11
- CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques — tomssilver · 2026-11-11
- CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques — tomssilver · 2026-11-11
- CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques — tomssilver · 2026-11-11
- CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques — tomssilver · 2026-11-11
- Formal verification could make AI coding faster: Lean proofs as agent guardrails — bigdata · 2026-11-11