Reproduce “ProMo3D: Probing Motion Cues in Frozen 3D Foundation 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
Xiaoang Zhang, Mert Kiray, Yordanka Velikova, Manoj Biswanath, Benjamin Busam
ProMo3D uses a supervised low-rank probe over frozen 3D-model tokens to segment moving regions despite camera motion. The probe compares token pairs and turns compatibility with a fixed reference into a coarse mask. On DAVIS17, coarse J-M/J-R scores are 73.7/91.0, above VGGT4D’s reported results, but below GeoMotion in J-M. Its annotation-defined masks need not reflect instantaneous physical motion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.