核心概念
Temporal equivariance and semantic continuity enhance weakly supervised surgical instrument segmentation.
要約
1. Introduction
Automated visual comprehension in laparoscopic and robotic surgery is crucial.
Surgical instrument segmentation (SIS) is foundational, evolving from traditional to deep learning methods.
Challenges in SIS include robustness and precision for clinical needs.
2. Methodologies
Unsupervised, semi-supervised, and weakly supervised paradigms explored for SIS.
WeakSurg introduces a novel weakly supervised architecture for SIS with temporal attributes.
3. Experiments
Conducted on Cholec80 dataset with instance-wise annotations by a clinician.
Results show improvement over state-of-the-art methods in both semantic and instance segmentation metrics.
4. Conclusions
WeakSurg extends two-stage WSSS methods with prototype-based temporal equivariance regulation and class-aware temporal semantic continuity.
Extensive experiments validate the effectiveness of the proposed method.
統計
"Extensive experiments are validated on Cholec80."
"Our results show that WeakSurg compares favorably with state-of-the-art methods."