LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents

2026-03-06

AI Summary

Existing 3D mesh generation methods struggle to simultaneously achieve topological fidelity, geometric detail, and inference efficiency. This work proposes LATO, a topology-preserving implicit generative framework that uniquely integrates flow matching with a topology-aware sparse voxel-based implicit representation. LATO constructs a structured latent space via a sparse voxel VAE and directly predicts vertex positions and connectivity of explicit meshes through a vertex displacement field, thereby circumventing the need for iso-surface extraction. A two-stage flow matching mechanism is introduced to ensure both complex geometry and correct topology while significantly accelerating generation. Experiments demonstrate that LATO outperforms current approaches—based on iso-surface extraction, triangle diffusion, or autoregressive modeling—in both generation quality and speed.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

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.

Problem

• 3D mesh generation
• topology preservation
• explicit surface representation
• generative modeling
• mesh synthesis

Innovation

• topology-preserving latent
• Vertex Displacement Field
• sparse voxel VAE
• flow matching
• mesh connectivity prediction