Research map / 3D vision / Face and shape
DiffSurf: A Transformer-based Diffusion Model for Generating and Reconstructing 3D Surfaces in Pose
Yusuke Yoshiyasu, Leyuan Sun
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 |
|---|---|---|---|
| replace transformer layers with MLP layers in the diffusion network (ablation) | DiffSurf · SURREAL unconditional generation | 1-NNA: 55.6 -> 92.6 | Hurt |
| add long skip connections to the diffusion transformer (ablation) | DiffSurf · SURREAL unconditional generation | 1-NNA: 54.4 -> 55.6 | Hurt |
| use 3D template position embedding instead of learned (ablation) | DiffSurf · SURREAL unconditional generation | 1-NNA: 55.6 -> 54.8 | No clear effect |
| add time embedding instead of using a time token (ablation) | DiffSurf · SURREAL unconditional generation | 1-NNA: 55.6 -> 55.4 | No clear effect |
| increase DDIM sampling steps from 1 to 10-20 (ablation) | DiffSurf human mesh recovery · 3DPW | PA-MPJPE: 143.9 -> 54.3 (10 steps) | Improved |
| increase sampling steps from 10 to 100 (ablation) | DiffSurf human mesh recovery · 3DPW | PA-MPJPE / fps: 54.3 -> 54.7; fps 21.2 -> 3.38 | Hurt |
| increase CFG scale from 0 to 1.0 (ablation) | DiffSurf human mesh recovery · 3DPW | PA-MPJPE: 52.6 -> 54.3 | Hurt |
| increase CFG scale from 0 to 0.5 (ablation) | DiffSurf human mesh recovery · Human3.6M | PA-MPJPE: 37.9 -> 36.1 | Improved |
| add SDS-based fitting using DiffSurf prior to HMR-EFT with GT 2D keypoints | HMR-EFT · 3DPW fitting | PA-MPJPE: 54.3 -> 52.1 | Improved |
| add SDS-based fitting using DiffSurf prior to METRO | METRO · 3DPW fitting | PA-MPJPE: 63.0 -> 54.2 | Improved |
| add L_consist loss in control point deformation (ablation) | DiffSurf control point deformation · qualitative control point deformation | fit and distortion | Mixed |
| add edge and Laplacian SDS losses in control point deformation (ablation) | DiffSurf control point deformation · qualitative control point deformation | distortion | Improved |
Closest papers by meaning
- DiffHuman: Probabilistic Photorealistic 3D Reconstruction of Humans (CVPR 2024 · similarity 0.90)
- Coherent 3D Scene Diffusion From a Single RGB Image (NeurIPS 2024 · similarity 0.88)
- DMV3D: Denoising Multi-view Diffusion Using 3D Large Reconstruction Model (ICLR 2024 · similarity 0.88)
- Denoising Functional Maps: Diffusion Models for Shape Correspondence (CVPR 2025 · similarity 0.87)
- DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis (CVPR 2024 · similarity 0.87)
- SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · similarity 0.87)
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.
