CoGE: Online Geometric Estimation for Monocular Colonoscopy
Researchers present CoGE, a framework for online monocular geometric estimation during colonoscopy, addressing challenges of depth estimation and scene reconstruction in the narrow, enclosed colon environment. The framework includes an illumination-aware supervision module based on Retinex theory to handle diverse lighting conditions, and a structure-aware perception module using wavelet decomposition to extract common structural and local features. Trained solely on simulated data, CoGE achieves strong quantitative and qualitative results, bridging the feature gap between simulation and real-world artifacts and illumination. The work is detailed in arXiv:2605.13038.
Key facts
- CoGE is a framework for online monocular geometric estimation during colonoscopy.
- It includes an illumination-aware supervision module based on Retinex theory.
- A structure-aware perception module uses wavelet decomposition.
- The model is trained solely on simulated data.
- It addresses depth estimation and scene reconstruction in colonoscopy.
- The framework handles illumination diversity and feature gaps between simulated and real data.
- Quantitative and qualitative results demonstrate effectiveness.
- The paper is available on arXiv with ID 2605.13038.
Entities
Institutions
- arXiv