AI
SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 30/09/2026, 20:00:00 UTC
Briefing created: 03/10/2026, 01:46:58 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 05:18:24 UTC.
Briefings
SILSA is a topology-aware 3D generation framework that replaces expensive voxel tokens with compact sliding-window slice latents. By using a fixed set of overlapping slices along the three canonical axes, SILSA preserves cross-sectional continuity and enables single-stage rectified-flow generation. The framework introduces slice-level topology supervision to match persistence diagrams and align Betti transitions across neighboring slices. Experiments demonstrate that SILSA improves structural fidelity by reducing generation cost, achieving an 8.7% PSNR gain, 5.96 absolute coverage points, and a 9.2% reduction in Betti error compared to the strongest baseline. It uses 70.0% fewer tokens than the next-most compact baseline and over 98% fewer tokens than sparse or hierarchical tokenizers, resulting in 40.4% lower training memory and 58.5% faster inference time. Qualitative results show improved preservation of thin structures, repeated components, and long-range connectivity.
Key points
- SILSA uses a fixed set of overlapping slices along the three canonical axes to represent shapes with compact sliding-window slice latents instead of expensive voxel tokens.
- Slice-level topology supervision matches persistence diagrams and aligns Betti transitions across neighboring slices to preserve structural correctness.
Context and limitations
The study compares SILSA against baselines using only PSNR, coverage, and Betti error metrics; qualitative improvements in thin structures and long-range connectivity are observed but not quantified numerically.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 03/10/2026, 01:46:58 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifnfg6jvo4lasz5odjxey5hbcofahehpawfzyouphiyxbdybyttru