Reproduce “DiffusionDepth: Diffusion Denoising Approach for Monocular Depth Estimation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiqun Duan, Xianda Guo, Zheng Zhu
Researchers from UTS and PhiGent Robotics introduce DiffusionDepth, the first work to apply diffusion models to monocular depth estimation, achieving state-of-the-art results on KITTI and NYU-Depth-V2 by generating highly detailed and accurate depth maps, even with sparse ground truth data.
Key insights
- Diffusion models can be successfully adapted for discriminative 3D perception tasks, extending their utility beyond generative image synthesis.
- The iterative denoising process inherently supports progressive refinement, allowing the model to capture both coarse scene structures and fine-grained details for high-quality depth estimation.
- The self-diffusion mechanism is crucial for training diffusion-based MDE models effectively on sparse ground truth data, preventing mode collapse and ensuring comprehensive depth prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.