SymTRELLIS: Symmetry-Enforced Voxel Latents for 3D Generation

2026-06-02

AI Summary

This work addresses the structural and functional inadequacies often produced by single-view 3D generation models due to their neglect of symmetry. Building upon the TRELLIS.2 streaming 3D generation framework, the authors propose a method to dynamically enforce arbitrary finite point group symmetries—such as rotational, reflectional, and polyhedral symmetries—without retraining the underlying VAE or flow model. The approach learns linear operators in latent space that represent spatial transformations and applies symmetric averaging to the velocity field at every step of the ODE solver, explicitly imposing the desired symmetries. It supports both automatic and user-specified symmetry types, significantly reducing symmetry errors on a benchmark of 266 strictly symmetric objects while preserving reconstruction accuracy comparable to the original model, outperforming existing methods such as Hunyuan3D-2.1 and TripoSG.

Abstract

Single-view 3D generative models have achieved impressive visual quality, yet they are not designed to satisfy structural or functional requirements, and in practice, often fall short. Symmetry is one such requirement: violations, even subtle ones, on symmetry can render a model physically unusable. We present SymTRELLIS, a method that enforces arbitrary finite point group symmetries (rotational, reflectional, and polyhedral) during the flow-based 3D generation of TRELLIS.2, without retraining the underlying VAE or flow model. Our key idea is to approximate the latent-space action of spatial transformations as a learned linear operator on voxel latents, implemented as a lightweight spatial-transform latent mapper trained on generic, non-symmetric 3D data. At generation time, we enforce symmetry by averaging predicted flow velocities across all symmetry-equivalent transformations at each ODE step, a process we call velocity symmetrization. The symmetry specification can be estimated automatically from an initial TRELLIS.2 generation or supplied by the user, enabling deliberate fold manipulation beyond what the input image suggests. On a curated benchmark of 266 strictly symmetric objects spanning 2- to 20-fold rotations and polyhedral symmetry groups, SymTRELLIS substantially reduces all symmetry error metrics compared to TRELLIS.2, Hunyuan3D-2.1, and TripoSG, while maintaining reconstruction accuracy comparable to the base model.

Problem

• 3D generation
• symmetry
• single-view reconstruction
• structural constraints
• point group symmetry

Innovation

• symmetry enforcement
• latent-space transformation
• flow-based 3D generation
• velocity symmetrization
• point group symmetry