Research map / 3D vision / Face and shape
TriFlow: Generating Artist-Like 3D Mesh Topology via Nearest-Vertex Vector Fields
Haoxuan Li, Ziya Erkoç, Daniele Sirigatti, Vladislav Rosov, Lei Li, Angela Dai, Matthias Niessner
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 watershed grouping (ablation) | TriFlow · TRELLIS-generated shapes | FID: 25.95 -> 16.19 | Improved |
| add watershed grouping (ablation) | TriFlow · TRELLIS-generated shapes | VLM topology score: 3.8 -> 4.0 | Improved |
| add constrained QEM instead of vertex flow (ablation) | TriFlow · TRELLIS-generated shapes | FID: 29.13 -> 16.19 | Improved |
| add constrained QEM instead of vertex flow (ablation) | TriFlow · TRELLIS-generated shapes | VLM geometry score: 4.1 -> 4.4 | Improved |
| add geometric distortion augmentation (ablation) | TriFlow · TRELLIS-generated shapes | FID: 18.06 -> 16.19 | Improved |
| add geometric distortion augmentation (ablation) | TriFlow · TRELLIS-generated shapes | Chamfer Distance (x1000): 0.20 -> 0.20 | No clear effect |
| use direction and sqrt-length parameterization instead of raw vectors (ablation) | NVF VAE/flow · NVF parameterization (128^3) | directional cosine similarity: 0.17 -> 0.26 | Improved |
| use direction and sqrt-length parameterization instead of target positions (ablation) | NVF VAE/flow · NVF parameterization (128^3) | l2 distance of NVF: 0.16 -> 0.10 | Improved |
| replace autoregressive baselines with TriFlow | TreeMeshGPT · Objaverse | Chamfer Distance (x1000): 0.98 -> 0.12 | Improved |
| replace autoregressive baselines with TriFlow on out-of-distribution shapes | TreeMeshGPT · TRELLIS-generated shapes | Chamfer Distance (x1000): 30.00 -> 0.20 | Improved |
| replace TreeMeshGPT by TriFlow | TreeMeshGPT · runtime | time per sample: 4.3 minutes -> 31 seconds | Improved |
| replace QEM by TriFlow | QEM · TRELLIS-generated shapes | FID: 22.1 -> 16.2 | Improved |
Closest papers by meaning
- MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers (CVPR 2024 · similarity 0.89)
- MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers (ICLR 2025 · similarity 0.88)
- SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation (NeurIPS 2026 · similarity 0.87)
- CraftsMan3D: High-fidelity Mesh Generation with 3D Native Diffusion and Interactive Geometry Refiner (CVPR 2025 · similarity 0.87)
- MeshArt: Generating Articulated Meshes with Structure-Guided Transformers (CVPR 2025 · similarity 0.86)
- Flow-Corrected Shape Optimization: Taming Manifold Drift in High-Dimensional 3D Models (NeurIPS 2026 · 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.
