EcoDepth
ECoDepth is the official implementation of the CVPR 2024 paper counting the Effective Conditioning of Diffusion Models for Monocular Depth Estimation. It provides a state-of-the-art framework for generating accurate depth maps from single images using diffusion models. The software features a modernized codebase architecture that streamlines training, testing, and inference workflows through PyTorch Lightning integration. Key capabilities include automatic model checkpoint management with built-in caching, a flexible DepthDataset module for custom dataset loading, and a configuration-driven approach using JSON files instead of complex bash scripts. Major improvements in version 2.0 reduce dependency overhead by eliminating the need for manual installation of full Stable Diffusion or CLIP pipelines, as required weights are embedded. The system supports comprehensive workflows including training on NYUv2 and KITTI datasets, evaluating model performance, and performing inference on both static images and video s