Reproduce “ScenePlacer: Reformulating Scene-Level Object Pose Estimation as Visibility Field Learning” 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
Ken Deng, Jiarui Liu, Guibiao Liao, Kunming Luo, Mochu Xiang, Hongyu Yan, Yebin Liu, Shenghua Gao and 2 more
ScenePlacer separates object generation from placement by predicting which parts of a generated mesh are visible in an observation, then aligning those parts to partial points. On 1,000 3D-FUTURE scenes with MoGe depth and SAM3D objects, it achieves Chamfer Distance 0.77 versus 1.31 and F-score 92.40 versus 88.77; accuracy remains dependent on input geometry.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.