Reproduce “LISA-3D: Lifting Language-Image Segmentation to 3D via Multi-View Consistency” 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
Zhongbin Guo, Jiahe Liu, Wenyu Gao, Yushan Li, Xiaomin He, Chengzhi Li, Ping Jian
LISA-3D adapts LISA’s masks with geometric supervision from registered RGB-D views and camera poses, while keeping the LISA backbone and SAM-3D frozen. On ScanRefer, its one-test-view results are mIoU 17.6, surface F-score 61.8, and CD 10.2, versus 10.2, 54.7, and 12.4 for LISA with SAM-3D. This benefit follows paired-view training; a controlled ablation supports added value from reprojection loss.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.