SILSA: Topology-Preserving High-Resolution 3D Generation with Sliding-Window Slice Latents
SILSA：利用滑動視窗切片潛在特徵實現保持拓撲結構的高解析度 3D 生成
High-resolution 3D generation often struggles with fragmented surfaces and high computational costs caused by voxel representation. SILSA bypasses voxel tokens by utilizing a fixed set of overlapping sliding-window slices across three canonical axes. Integrating a Slice VAE, a Volumetric Anchor Lattice, and slice-level topology supervision (aligning Betti transitions), SILSA enables efficient single-stage rectified-flow 3D generation with outstanding structural fidelity and minimal resource overhead.
Key points
Sliding-Window Slice Latents
Uses overlapping sliding-window slices along three axes to project local depth into 2D latents, preserving cross-sectional continuity.
Sparse Decoder & Shared Workspace
Employs a Slice VAE and a Volumetric Anchor Lattice to coordinate multi-axis slice streams within a shared 3D workspace.
Slice-Level Topology Supervision
Integrates persistence diagram matching and Betti transition alignment to prevent thin or complex connections from breaking.
Unmatched Token Efficiency
Requires 70% fewer tokens than the next-most compact baseline, cutting training memory by 40.4% and inference time by 58.5%.
How it works
Why it matters
Generating topologically accurate thin structures (such as wire fences or chair legs) has been a bottleneck in 3D asset creation due to high voxel computational requirements. SILSA demonstrates that multi-axis 2D slice projection, paired with topology guidance, can reconstruct high-fidelity 3D shapes at a fraction of the cost. This opens up highly efficient, single-stage 3D generation on consumer-grade hardware for gaming, VR, and simulation.
Who it affects
- AI Researcher
- AI Developer
- Content Creator
How to use it
- 1High-resolution 3D asset generation for gaming, especially for complex shapes containing thin, repetitive, or porous elements.
- 2Creating geometric digital twins and robot simulation environments with minimal memory and compute overhead.
Limitations & caveats
- The model relies on predefined orthogonal slicing axes, which may limit performance on highly asymmetric, tilted, or non-manifold geometric shapes.
- Calculating persistent homology for slice-level topology supervision is computationally demanding and may cause CPU bottlenecks during massive-scale training.
Related
Ai2 Open-Sources AstaBrief: An 8B Scientific Report Generator 3.5x Faster than Claude
艾倫人工智慧研究所開源 AstaBrief：比 Claude 快 3.5 倍的 8B 科學報告生成模型
Allen Institute for AI (Ai2) has open-sourced AstaBrief 8B, a specialized model for scientific report generation that achieves a 3.5x speedup over proprietary pipelines while maintaining high citation accuracy.
GALA: Distilling 3D Gaussian Avatars into Linear Blendshapes for Real-Time Animation
GALA：用線性混合變形蒸餾技術實現 3D Gaussian 虛擬化身即時動畫
GALA distills complex neural decoding of 3D Gaussian avatars into lightweight linear blendshapes, reducing CPU animation costs by up to 1000x and enabling 60fps real-time performance on mobile devices.
ScholarCatalyst: A Benchmark for Testing AI's Intuition in Retrieving Inspiring Research Papers
ScholarCatalyst：評估 AI 是否擁有「科學家直覺」的學術文獻檢索基準
ScholarCatalyst is a novel benchmark featuring annotations from 184 lead authors to evaluate whether AI can retrieve key inspiring papers from past literature based only on an initial research question.