Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.06357 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912948524417024 |
|---|---|
| author | Zhao, Tianhao Zhang, Youjia Long, Hang Zhang, Jinshen Li, Wenbing Yang, Yang Zhang, Gongbo Hladký, Jozef Nießner, Matthias Yang, Wei |
| author_facet | Zhao, Tianhao Zhang, Youjia Long, Hang Zhang, Jinshen Li, Wenbing Yang, Yang Zhang, Gongbo Hladký, Jozef Nießner, Matthias Yang, Wei |
| contents | In this paper, we introduce LATO, a novel topology-preserving latent representation that enables scalable, flow matching-based synthesis of explicit 3D meshes. LATO represents a mesh as a Vertex Displacement Field (VDF) anchored on surface, incorporating a sparse voxel Variational Autoencoder (VAE) to compress this explicit signal into a structured, topology-aware voxel latent. To decapsulate the mesh, the VAE decoder progressively subdivides and prunes latent voxels to instantiate precise vertex locations. In the end, a dedicated connection head queries the voxel latent to predict edge connectivity between vertex pairs directly, allowing mesh topology to be recovered without isosurface extraction or heuristic meshing. For generative modeling, LATO adopts a two-stage flow matching process, first synthesizing the structure voxels and subsequently refining the voxel-wise topology features. Compared to prior isosurface/triangle-based diffusion models and autoregressive generation approaches, LATO generates meshes with complex geometry, well-formed topology while being highly efficient in inference. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_06357 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents Zhao, Tianhao Zhang, Youjia Long, Hang Zhang, Jinshen Li, Wenbing Yang, Yang Zhang, Gongbo Hladký, Jozef Nießner, Matthias Yang, Wei Computer Vision and Pattern Recognition In this paper, we introduce LATO, a novel topology-preserving latent representation that enables scalable, flow matching-based synthesis of explicit 3D meshes. LATO represents a mesh as a Vertex Displacement Field (VDF) anchored on surface, incorporating a sparse voxel Variational Autoencoder (VAE) to compress this explicit signal into a structured, topology-aware voxel latent. To decapsulate the mesh, the VAE decoder progressively subdivides and prunes latent voxels to instantiate precise vertex locations. In the end, a dedicated connection head queries the voxel latent to predict edge connectivity between vertex pairs directly, allowing mesh topology to be recovered without isosurface extraction or heuristic meshing. For generative modeling, LATO adopts a two-stage flow matching process, first synthesizing the structure voxels and subsequently refining the voxel-wise topology features. Compared to prior isosurface/triangle-based diffusion models and autoregressive generation approaches, LATO generates meshes with complex geometry, well-formed topology while being highly efficient in inference. |
| title | LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.06357 |