Geometry-Consistent Endoscopic Representations for Image-Guided Navigation via Structured Foundation Model Adaptation

The article presents a unified framework for enhancing geometry-consistent representations in monocular endoscopy, addressing challenges in pose estimation and depth prediction. It introduces Hierarchy-Aware Geometry-Semantic Adaptation, which employs selective low-rank adapters within the transformer architecture and integrates geometric supervision from synthetic data. Experimental results demonstrate improved representation quality and performance in navigation tasks, indicating that this approach could significantly aid practitioners in developing more reliable endoscopic navigation systems.