Research map / 3D vision / 3D generation
DiffuStereo: High Quality Human Reconstruction via Diffusion-Based Stereo Using Sparse Cameras
Ruizhi Shao, Zerong Zheng, Hongwen Zhang, Jingxiang Sun, Yebin Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove diffusion model from the stereo network (add diffusion) (ablation) | DiffuStereo stereo network · THuman2.0 stereo (45 deg) | average disparity error (pixels): 1.448 -> 0.632 | Improved |
| use the proposed diffusion kernel instead of the original 1000-step kernel (ablation) | DiffuStereo · THuman2.0 stereo (45 deg) | average disparity error (pixels): 0.881 -> 0.632 | Improved |
| use the proposed diffusion kernel instead of the original kernel (ablation) | DiffuStereo · THuman2.0 stereo (45 deg) | disparity error within 1/2 pixel (%): 52.9 -> 67.6 | Improved |
| add diffusion-based stereo refinement on top of RAFT-Stereo given the same coarse depth Dc | RAFT-Stereo (w. Dc) · THuman2.0 stereo (45 deg) | average disparity error (pixels): 1.567 -> 0.632 | Improved |
| use the proposed diffusion kernel (original kernel unstable) (ablation) | DiffuStereo · depth maps | depth map stability | Improved |
| reduce reverse diffusion to about 10 steps with the new kernel (ablation) | DiffuStereo · depth maps | detail recovery | No clear effect |
| use the proposed 30-step diffusion instead of the original 1000-step model (ablation) | DiffuStereo · inference time | time: 12s with 30 diffusion steps | Improved |
| use visual hull from 8 views as the initial coarse model instead of DoubleField mesh (ablation) | DiffuStereo · coarse model initialization | visible-region detail | No clear effect |
| increase view angle between stereo pair from 20 to 45 deg | DiffuStereo · THuman2.0 stereo | disparity error within 1/2 pixel (%): 71.3 -> 67.6 | Hurt |
| replace DoubleField reconstruction with DiffuStereo refinement | DoubleField · THuman2.0 reconstruction | Chamfer distance (mm): 2.879 -> 1.198 | Improved |
| replace DoubleField reconstruction with DiffuStereo refinement | DoubleField · THuman2.0 reconstruction | ratio within 1mm (%): 23.2 -> 68.1 | Improved |
Closest papers by meaning
- DiffHuman: Probabilistic Photorealistic 3D Reconstruction of Humans (CVPR 2024 · similarity 0.89)
- Mono2Stereo: A Benchmark and Empirical Study for Stereo Conversion (CVPR 2025 · similarity 0.87)
- Coherent 3D Scene Diffusion From a Single RGB Image (NeurIPS 2024 · similarity 0.87)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.87)
- Adaptive 3D Reconstruction via Diffusion Priors and Forward Curvature-Matching Likelihood Updates (NeurIPS 2025 · similarity 0.86)
- Zero-Shot Novel View and Depth Synthesis with Multi-View Geometric Diffusion (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
