Reproduce “FILIGREE3D: Scaling Sparse Latent Flow Matching for Ultra-High-Resolution Image-to-3D Generation” 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
Hongjie Li, Xinran Yang, Xiuchao Wu, Jiangjing Lyu, Chengfei Lv
Filigree3D reconstructs detailed 3D geometry from a single image by combining bounded local attention on sparse representations with periodic global attention over pooled coarse tokens. It uses adaptive crops, multi-scale image features, and core-region supervision to limit computation while preserving context. On FineGeo3D, its 2048³ model achieves strict F1-0.002 of 0.4309 versus 0.0674 for baselines; 4096³ is qualitative only.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.