Research map / 3D vision / Face and shape
3D-ReGen: A Unified 3D Geometry Regeneration Framework
Geon Yeong Park, Roman Shapovalov, Rakesh Ranjan, Jong Chul Ye, Andrea Vedaldi, Thu Nguyen-Phuoc
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 |
|---|---|---|---|
| regenerate with MeshReGen instead of leaving input shape | coarse decomposed objects · 3D shape enhancement | MV-ImageReward: 0.1716 -> 0.3394 | Improved |
| regenerate with MeshReGen instead of DetailGen3D | coarse decomposed objects · 3D shape enhancement | ULIP-3D: 0.2294 -> 0.2626 | Improved |
| add MeshReGen refinement to VGGT predictions (4 views) | VGGT · GSO reconstruction | Chamfer distance: 0.1276 -> 0.0081 | Improved |
| add MeshReGen refinement to VGGT predictions (1 view) | VGGT · GSO reconstruction | Chamfer distance: 0.1286 -> 0.0135 | Improved |
| more input views for MeshReGen (1 to 4) | VGGT + MeshReGen · GSO reconstruction | F-score: 0.3658 -> 0.4913 | Improved |
| MeshReGen with 4 views versus AssetGen on PSNR | VGGT + MeshReGen · GSO reconstruction (4 views) | PSNR: 24.3809 -> 24.2754 | No clear effect |
| use MeshReGen instead of Hunyuan3D-Omni (1 view) | GSO reconstruction · GSO reconstruction (1 view) | Chamfer distance: 0.0173 -> 0.0135 | Improved |
| use input-space conditioning instead of additional cross-attention (ablation) | MeshReGen · Conditioning ablation | Chamfer distance: 0.0066 -> 0.0044 | Improved |
| use input-space conditioning instead of extended cross-attention (ablation) | MeshReGen · Conditioning ablation | Chamfer distance: 0.0271 -> 0.0044 | Improved |
| use coarse-shape embedding conditioning instead of additively injecting zL into noisy latents (ablation) | MeshReGen · Conditioning ablation | F-score: 0.5259 -> 0.7538 | Improved |
| condition on sampled point clouds instead of coarse embeddings in cross-attention (ablation) | MeshReGen · Conditioning ablation | Chamfer distance: 0.0066 -> 0.0162 | Hurt |
| keep original diffusion parameterization instead of flow mapping from coarse to clean latents (ablation) | MeshReGen · Diffusion parameterization ablation | F-score: 0.6026 -> 0.7003 | Improved |
Closest papers by meaning
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.89)
- ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction (NeurIPS 2025 · similarity 0.89)
- MeshGen: Generating PBR Textured Mesh with Render-Enhanced Auto-Encoder and Generative Data Augmentation (CVPR 2025 · similarity 0.89)
- MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers (ICLR 2025 · similarity 0.88)
- MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers (CVPR 2024 · similarity 0.88)
- MeshFormer : High-Quality Mesh Generation with 3D-Guided Reconstruction Model (NeurIPS 2024 · similarity 0.88)
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.
