SF3D: Stable Fast 3D Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement


SF3D takes a single image as input and generates a textured UV-unwrapped 3D model in under a second.

Abstract

We present Stable Fast 3D (SF3D) - a feed-forward 3D asset generation model , which generates UV-unwrapped, textured models with material parameters and delit albedo in under a second.

Utilizing the Large Reconstruction Model (LRM) as its foundation, SF3D achieves impressive performance, reconstructing detailed, textured meshes in just 0.5 seconds. Unlike traditional approaches, SF3D is explicitly trained for mesh generation, incorporating a fast UV unwrapping technique that enables swift texture generation rather than relying on vertex colors. Additionally, the method learns material parameters and normal maps to enhance the visual quality of the reconstructed models. Furthermore, SF3D integrates a delighting step to effectively remove low-frequency illumination effects, ensuring that the reconstructed meshes can be easily used in novel illumination conditions.

Summary Video

Comparison and Results

Asset Generation

Results on a set of diverse images from GSO (License), OmniObject3D (License), and we also generate some with Stable Assistant.

Comparing our results with other state-of-the-art image-to-3D methods. Image taken from GSO.

Loading...

SF3D (ours)

Loading...

TripoSR

Loading...

CRM

Loading...

InstantMesh


Click on the individual images to view the 3D model. Input images taken from GSO, OmniObject3D, and we also generate some with Stable Assistant.

    BibTeX

    @article{boss2024sf3d,
      author    = {Boss, Mark and Huang, Zixuan and Vasishta, Aaryaman and Jampani, Varun},
      title     = {{SF3D}: Stable Fast {3D} Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement},
      journal   = {{arXiv}},
      year      = {2024},
    }