Reproduce “MeshRipple: Structured Autoregressive Generation of Artist-Meshes” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junkai Lin, Hang Long, Huipeng Guo, Jielei Zhang, JiaYi Yang, Tianle Guo, Yang Yang, Jianwen Li and 3 more
MeshRipple introduces a structured autoregressive framework to overcome the topology-context mismatch in 3D mesh generation, enabling the creation of high-fidelity, topologically coherent meshes. The method achieves superior performance on artist-mesh benchmarks (Chamfer Distance: 46.68, Hausdorff Distance: 0.0938, Normal Consistency: 0.5166) with improved computational efficiency compared to prior approaches.
Key insights
- Aligning the mesh tokenization order with its topological expansion (via a frontier-aware BFS) inherently resolves the critical context mismatch problem in truncated autoregressive training.
- Explicitly guiding the mesh growth by predicting both the next face and the specific next expansion point on the frontier significantly enhances topological coherence and reduces fragmentation.
- Combining a local, topology-aware context with a memory-efficient global context mechanism (NSCA) allows the autoregressive model to scale to large, complex meshes while maintaining structural integrity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.