PhysGen: Physically Grounded 3D Shape Generation for Industrial Design

TL;DR

PhysGen introduces a physics-guided 3D shape generation method that integrates physical properties into the generative process, enhancing realism for industrial design applications.

Contribution

A novel physics-based 3D shape generation pipeline using a flow matching model and a shape-and-physics VAE, enabling physically valid and realistic shape synthesis.

Findings

Improved shape realism over existing models.

Effective incorporation of physical properties into shape generation.

Validated on three benchmark datasets.

Abstract

Existing generative models for 3D shapes can synthesize high-fidelity and visually plausible shapes. For certain classes of shapes that have undergone an engineering design process, the realism of the shape is tightly coupled with the underlying physical properties, e.g., aerodynamic efficiency for automobiles. Since existing methods lack knowledge of such physics, they are unable to use this knowledge to enhance the realism of shape generation. Motivated by this, we propose a unified physics-based 3D shape generation pipeline, with a focus on industrial design applications. Specifically, we introduce a new flow matching model with explicit physical guidance, consisting of an alternating update process. We iteratively perform a velocity-based update and a physics-based refinement, progressively adjusting the latent code to align with the desired 3D shapes and physical properties. We further strengthen physical validity by incorporating a physics-aware regularization term into the velocity-based update step. To support such physics-guided updates, we build a shape-and-physics variational autoencoder (SP-VAE) that jointly encodes shape and physics information into a unified latent space. The experiments on three benchmarks show that this synergistic formulation improves shape realism beyond mere visual plausibility. Our code and model weights are available at https://github.com/kasvii/PhysGen.