PhySe-RPO: AI Model for Surgical Smoke Removal in Video
A team of researchers has introduced PhySe-RPO, a diffusion-based system designed to eliminate surgical smoke from intraoperative videos. This model shifts from deterministic restoration to a stochastic policy, allowing for exploration at the trajectory level and updates that do not require a critic through group-relative optimization. It incorporates a physics-informed reward to ensure consistency in illumination and color, alongside a CLIP-based semantic reward for restoring anatomically coherent, smoke-free visuals. This new method overcomes the challenges faced by current learning-based desmoking techniques that depend on limited paired supervision and deterministic processes. The findings are published in a paper available on arXiv (ID: 2603.22844).
Key facts
- PhySe-RPO is a diffusion restoration framework for surgical smoke removal.
- It uses Physics- and Semantics-Guided Relative Policy Optimization.
- The method transforms deterministic restoration into a stochastic policy.
- A physics-guided reward ensures illumination and color consistency.
- A CLIP-based semantic reward promotes smoke-free restoration.
- The approach enables trajectory-level exploration and critic-free updates.
- Existing methods rely on scarce paired supervision and deterministic pipelines.
- The paper is available on arXiv with ID 2603.22844.
Entities
Institutions
- arXiv