WEDepth: Efficient Adaptation of World Knowledge for Monocular Depth Estimation

The introduction of WEDepth marks a significant advancement in monocular depth estimation (MDE), a challenging area in computer vision due to the difficulty of reconstructing 3D scenes from 2D images. By leveraging Vision Foundation Models (VFMs) without modifying their structures, WEDepth enhances MDE performance through systematic prior knowledge injection. Experiments conducted on the NYU-Depth v2 and KITTI datasets demonstrate that WEDepth achieves state-of-the-art performance, outperforming traditional diffusion-based methods and those pre-trained on relative depth. Additionally, its strong zero-shot transfer capability suggests that WEDepth can be effectively applied across various vision tasks, underscoring its importance in advancing the field of computer vision.