Research map / 3D vision / Face and shape
LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents
Tianhao Zhao, Youjia Zhang, Hang Long, Jinshen Zhang, Wenbing Li, Yang Yang, Gongbo Zhang, Jozef Hladký, Matthias Nießner, Wei Yang
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 VDF displacement field to VAE input (ablation) | LATO VAE · VAE input ablation | HD: 0.092 -> 0.075 | Improved |
| add VDF displacement field to VAE input (ablation) | LATO VAE · VAE input ablation | CD(L2): 0.043 -> 0.039 | Improved |
| add normal vectors to VAE input (ablation) | LATO VAE · VAE input ablation | |NC|: 0.832 -> 0.837 | Improved |
| add normal vectors to VAE input (ablation) | LATO VAE · VAE input ablation | HD: 0.076 -> 0.075 | Improved |
| increase sampled surface points from 81,920 to 819,200 (ablation) | LATO VAE · VAE input ablation | HD: 0.078 -> 0.075 | Improved |
| normalize each displacement vector (ablation) | LATO VAE · VAE input ablation | CD(L2): 0.039 -> 0.042 | Hurt |
| normalize displacement only above threshold tau (ablation) | LATO VAE · VAE input ablation | CD(L2): 0.039 -> 0.041 | Hurt |
| use VDF instead of discrete voxel labels as VAE input (ablation) | LATO · VDF vs labeled voxels | qualitative structure | Improved |
| add T-Voxel cross-attention in decoder (ablation) | LATO decoder · T-Voxel cross-attention ablation | qualitative mesh | Improved |
| replace geometric connectivity algorithm with learnable Connection Head (ablation) | LATO · connectivity inference | generated sample quality | Improved |
| use continuous voxel VAE instead of per-face discrete tokens | LATO vs MeshGPT · VAE reconstruction (artistic meshes) | CD(L2): 0.050 -> 0.040 | Improved |
| use flow matching on T-Voxels instead of autoregressive generation | LATO vs DeepMesh · Geometry-conditioned generation (dense meshes) | CD(L2): 0.051 -> 0.043 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- MeshAnythingv2 — MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers (ICLR 2025) · better on 4 of 4 shared results
- DeepMesh — DeepMesh: Differentiable Iso-Surface Extraction (TPAMI 2024) · better on 4 of 4 shared results
- MeshGPT — MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers (CVPR 2024) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation (NeurIPS 2026 · similarity 0.87)
- MeshFormer : High-Quality Mesh Generation with 3D-Guided Reconstruction Model (NeurIPS 2024 · similarity 0.85)
- ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction (NeurIPS 2025 · similarity 0.85)
- Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modeling (NeurIPS 2025 · similarity 0.85)
- Flow-Corrected Shape Optimization: Taming Manifold Drift in High-Dimensional 3D Models (NeurIPS 2026 · similarity 0.85)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.85)
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.
