Research map / 3D vision / Face and shape
TripoSG: High-Fidelity 3D Shape Synthesis Using Large-Scale Rectified Flow Models
Yangguang Li, Zi–Xin Zou, Zexiang Liu, Dehu Wang, Yuan Liang, Zhipeng Yu, Xingchao Liu, Yuan–Chen Guo, Liang Ding, Wanli Ouyang, Yan‐Pei Cao
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 |
|---|---|---|---|
| add CLIP condition to DINOv2 (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 10.69 -> 10.61 | Improved |
| add skip-connection (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 10.61 -> 9.47 | Improved |
| use R-Flow sampling instead of DDPM (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 9.63 -> 9.47 | Improved |
| use R-Flow sampling instead of EDM (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 9.50 -> 9.47 | Improved |
| raise latent resolution from 512 to 2048 tokens (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 9.47 -> 8.38 | Improved |
| raise latent resolution from 2048 to 4096 tokens (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 8.38 -> 8.12 | Improved |
| add MoE at 4096 tokens (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 8.12 -> 7.94 | Improved |
| scale data from 180K Objaverse to 2M TripoSG data (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 9.47 -> 5.81 | Improved |
| apply data-building system (800K raw -> 180K filtered) (ablation) | flow model · Objaverse 180K ablation (Normal-FID) | Normal-FID: 11.61 -> 9.47 | Improved |
| use SDF instead of occupancy representation (ablation) | VAE · VAE reconstruction (Objaverse 180K) | normal consistency: 0.952 -> 0.955 | Improved |
| add surface normal loss to SDF VAE (ablation) | VAE · VAE reconstruction (Objaverse 180K) | Chamfer distance: 4.60 -> 4.56 | Improved |
| add eikonal regularization (ablation) | VAE · VAE reconstruction (Objaverse 180K) | normal consistency: 0.956 -> 0.957 | Improved |
Closest papers by meaning
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.88)
- Flow-Corrected Shape Optimization: Taming Manifold Drift in High-Dimensional 3D Models (NeurIPS 2026 · similarity 0.88)
- One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion (CVPR 2024 · similarity 0.88)
- Cycle3D: High-quality and Consistent Image-to-3D Generation via Generation-Reconstruction Cycle (AAAI 2025 · similarity 0.88)
- Make-A-Shape: a Ten-Million-scale 3D Shape Model (ICML 2024 · similarity 0.88)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · 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.
