ColonAdapter: Geometry Estimation Through Foundation Model Adaptation for Colonoscopy

2025-11-27

AI Summary

This work addresses inaccurate 3D geometric estimation in monocular colonoscopy images—caused by non-Lambertian reflectance, moving illumination, and large textureless regions. We propose a self-supervised fine-tuning framework to adapt general-purpose geometric foundation models for clinical settings, improving accuracy in depth prediction, camera pose estimation, and dense point-cloud reconstruction. Key innovations include a Detail Recovery Module (DRM), a geometric consistency loss, and a confidence-weighted photometric loss—enabling robust modeling of texture-poor regions and scale-consistent optimization without ground-truth camera intrinsics. Evaluated on both synthetic and real colonoscopy datasets, our method achieves state-of-the-art performance across monocular depth estimation, relative pose estimation, and 3D reconstruction tasks.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Estimating 3D geometry from monocular colonoscopy images is challenging due to non-Lambertian surfaces, moving light sources, and large textureless regions. While recent 3D geometric foundation models eliminate the need for multi-stage pipelines, their performance deteriorates in clinical scenes. These models are primarily trained on natural scene datasets and struggle with specularity and homogeneous textures typical in colonoscopy, leading to inaccurate geometry estimation. In this paper, we present ColonAdapter, a self-supervised fine-tuning framework that adapts geometric foundation models for colonoscopy geometry estimation. Our method leverages pretrained geometric priors while tailoring them to clinical data. To improve performance in low-texture regions and ensure scale consistency, we introduce a Detail Restoration Module (DRM) and a geometry consistency loss. Furthermore, a confidence-weighted photometric loss enhances training stability in clinical environments. Experiments on both synthetic and real datasets demonstrate that our approach achieves state-of-the-art performance in camera pose estimation, monocular depth prediction, and dense 3D point map reconstruction, without requiring ground-truth intrinsic parameters.

Problem

• Adapts geometric foundation models for colonoscopy geometry estimation
• Addresses challenges of specularity and homogeneous textures in colonoscopy
• Enhances accuracy in camera pose, depth prediction, and 3D reconstruction

Innovation

• Self-supervised fine-tuning of geometric foundation models
• Detail Restoration Module for low-texture region enhancement
• Confidence-weighted photometric loss for training stability