3DCarGen: Scalable 3D Car Generation via 3D-consistent Multi-view Synthesis

2026-06-23

AI Summary

Existing single-view 3D vehicle generation methods suffer from limited viewpoint coverage and cross-view geometric inconsistencies, resulting in low reconstruction fidelity. This work proposes a scalable generative framework that, for the first time, synthesizes an arbitrary number of geometrically consistent multi-view images from a single real-world input image by integrating explicit 3D priors into a diffusion model. The approach leverages a fast mesh reconstruction algorithm combining 3D Gaussian splatting with joint color-normal optimization, enabling high-fidelity geometry recovery. Evaluated on both synthetic and real-world datasets, the method significantly outperforms current state-of-the-art approaches, achieving detailed, view-coherent, and geometrically consistent 3D vehicle reconstructions.

Abstract

High-quality 3D vehicle assets are essential for autonomous driving simulation. Although multi-view diffusion-based paradigms enable controllable single-image reconstruction, they typically produce limited viewpoints and exhibit cross-view geometric inconsistencies, thereby reducing reconstruction fidelity in real-world scenarios. In this work, we introduce 3DCarGen, a scalable single-view 3D car generation framework designed for real-world images by synthesizing an arbitrary number of 3D-consistent multi-view images. Specifically, given a single image as input, we first synthesize a set of images from fixed viewpoints. These images are then fed into a feed-forward reconstruction model, resulting in a coarse 3D representation based on 3D Gaussian Splatting. Conditioned on this explicit 3D prior, our multi-view diffusion model generates 3D-consistent images from arbitrary camera viewpoints. We further extend a fast mesh reconstruction algorithm by incorporating color-normal joint optimization to recover detailed and coherent 3D vehicle models from the synthesized dense views. Extensive experiments on synthetic and real-world datasets demonstrate that our approach achieves robust geometric consistency and reconstruction fidelity compared to existing methods. Code and models will be released.

Problem

• 3D car generation
• multi-view synthesis
• geometric consistency
• 3D reconstruction
• autonomous driving simulation

Innovation

• 3D-consistent multi-view synthesis
• 3D Gaussian Splatting
• multi-view diffusion model
• color-normal joint optimization
• scalable 3D generation