[GitHub Trending] microsoft/TRELLIS.2
6.9 relevance
Score Breakdown
technical depth 9
novelty 9
actionability 3
community 7
strategic 6
personal 5
Scored daily by a customisable AI persona to surface the most relevant engineering leadership news.
Cutting-edge 3D generation model but not directly applicable to reader's workflows.
Summary
Microsoft's TRELLIS.2 introduces a 4B-parameter large 3D generative model using a field-free sparse voxel structure (O-Voxel) for high-fidelity image-to-3D generation. It leverages a Sparse 3D VAE with 16× spatial downsampling into compact latent space, enabling 512³ resolution textured assets with PBR materials (base color, roughness, metallic, opacity) in ~3 seconds on H100. The model handles arbitrary topologies including open surfaces and non-manifold geometry, with code and pretrained checkpoints available open-source.
Author
microsoft