Reproduce “Camera3R: Feed-Forward 3D Reconstruction from Heterogeneous Camera Models” 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
Ruxiao Duan, Yunwen Zhou, Erin Hong, Dongxu Zhao, Eric Turner, Federico Tombari, Alex Wong
Camera3R adapts perspective-pretrained multi-view reconstruction models to combine perspective, fisheye, and panoramic images in one feed-forward pass. It freezes the backbone, adds low-rank experts for non-perspective cameras, and routes each image to an expert while perspective images bypass adaptation. On mixed-camera KITTI360, π³ with Camera3R raises RRA/RTA/AUC scores from 0.200/0.082/0.010 to 1.000/0.881/0.753 versus unadapted π³, specifically on this test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.