UltraTex: Unleashing 2K Multi-View Diffusion for 3D Texturing
UltraTex enables 2048-resolution multi-view diffusion for 3D texturing with large training and inference speedups.
UltraTex is an end-to-end framework for 2048-resolution multi-view diffusion used to texture 3D assets. It drops background tokens before the diffusion transformer, applies block-sparse attention on the remaining foreground, and uses foreground-aware VAE decoding. Training uses G-buffer TexVerse, a dataset covering more than 268,000 ultra-high-resolution 3D assets. The paper reports about 20.6x to 91.1x training speedup and 22.3x to 74.6x end-to-end inference speedup versus the baseline.
- Targets 2048 resolution instead of 512 or 768
- Background token dropping and block-sparse attention cut cost
- G-buffer TexVerse covers over 268,000 3D assets
- Reports up to about 91x training and 75x inference speedups
From the original article
1 min read at huggingface.co
High-quality texture generation is essential for creating realistic and production-ready 3D assets. Recent multi-view diffusion methods have shown promising results for image-guided 3D texturing, but they are typically constrained to low operating resolutions such as 512 or 768, making it difficult to preserve high-frequency details from high-resolution reference images.
By Yibo Zhang, Ze Yuan, Nan Cao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.