SUPRA-TEX claims commercial-grade 3D from only public data
TL;DR
- SUPRA-TEX generates relightable 3D textures from a single reference image, with the authors stating the system was trained solely on public datasets.
- The method decouples visible-material estimation, surface alignment, and observation-guided 3D completion via what the authors call an interleaved flow transformer.
- The paper claims state-of-the-art results that match or exceed leading commercial systems on in-the-wild images, without naming specific competitors in the abstract.
SUPRA-TEX, a method for generating relightable 3D textures from a single reference image, claims to match or exceed leading commercial systems while training only on public datasets, according to a paper on arXiv from a team including Chen Yang and Jiemin Fang. The system takes a reference image together with a 3D mesh and outputs materials covering the full surface, aiming to "dress a given 3D mesh with relightable materials while faithfully preserving the visual details of a reference image."
The authors describe decoupling three steps: visible-material estimation, surface alignment, and observation-guided 3D completion. These are routed through what they call "an interleaved flow transformer" that holds the reference image as a fixed anchor while filling materials for unobserved surfaces. The paper reports "SOTA performance in both geometry-conditioned texturing and end-to-end asset generation, matching or even exceeding leading commercial systems on in-the-wild images."
The abstract names no specific commercial competitors and reports no per-metric scores.
Originally reported by paper
Read the original article →Original headline: Open-Source SUPRA-TEX Claims Commercial-Grade Single-Image 3D Texture on Public Data Alone