Active learning with weak supervision cuts surgical video annotation effort by ~50%
Laparoscopes · x · 2026-11-11
A University of Pennsylvania team published in IJCARS (presented at IPCAI) an operationalized active learning framework that speeds surgical video annotation by up to 50%.
- Method: a foundation model (DINOv3 features) generates temporally consistent class activation maps, trained with a weak-supervision loss on video-level tool-presence labels plus an image-level mask loss on human-corrected annotations.
- The pipeline iteratively proposes pseudo-masks that guide expert annotators to refine rather than densely label from scratch.
- By training's end, annotation effort dropped 50% versus fully manual labeling, enabling scalable surgical tool segmentation without large fully annotated datasets.
Related event: UPenn Team Speeds Up Surgical Video Annotation by 50% with Active Learning(3 posts)→
More from Research
- MacroHands: SOTA 3D hand tracking from egocentric video beats all open baselines — gui_penedo · 2026-11-12
- CypherGuard fixes structurally wrong LLM Cypher queries, hitting 88.5% on CypherBench — JeremyCMorgan · 2026-11-12
- Visualizing how Artificial Analysis Intelligence Index weights shift across every version — HamelHusain · 2026-11-12
- PufferLib 5.0 trains a Rubik's Cube solver at 8.8M SPS on a single 12GB GPU — haydendevs · 2026-11-11
- 10-line deterministic polynomial-time primality test relies on OpenAI's recent quasi-GRH result, making AKS obsolete — fortnow · 2026-11-11
- WSP Weight Edits Shown Equivalent to Constrained Activation Steering on Qwen — kastnerkyle · 2026-11-11