Reproduce “Mira-Scene: Pixel-Aligned Layouts for Generative 3D Scene Reconstruction” 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
Yang-Tian Sun, Tianjia Liu, Zehuan Huang, Yi-Hua Huang, Xiaoyang Lyu, Ziyi Yang, Zi-Xin Zou, Yuan-Chen Guo and 2 more
Mira-Scene reconstructs compositional 3D scenes from images by predicting each visible object pixel’s canonical surface coordinate and pairing it with a scene-space point map to recover placement. Its bounded canonical coordinate map is trained on object assets and adapted to scenes. On BlendSwap, 3D-IoU is 0.727 versus SAM3D’s 0.520, and 2D-IoU is 0.783 versus 0.672, under shared masks and alignment, not perfect reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.