Reproduce “Fuse3D: Generating 3D Assets Controlled by Multi-Image Fusion” 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
Xuancheng Jin, Rengan Xie, Wenting Zheng, Rui Wang, Hujun Bao, Yuchi Huo
Fuse3D combines whole-image context with selected regions from multiple references, using a pretrained 3D model’s attention to estimate correspondences between image tokens and voxels and regulate their influence during generation. In tested comparisons, it reports a region-specific CLIP score of 0.243 versus 0.240 for Blended Diffusion. These results support local and global image guidance, but alignment can be wrong.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.