Reproduce “3D-BBDA: Backpropagation-Free Black-Box Domain Adaptation for Robust 3D Point Cloud Recognition” 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
Shivangi Rai, Kunal Jangid, Vaibhav Kumar, Vinod K Kurmi
3D-BBDA adapts a frozen point-cloud classifier to unlabeled target data without source examples, source-model internals, or test-time gradients. It forms class prototypes from pseudo-labels, aligns target features to them, and blends alignment logits with source predictions. On ModelNet-40C with PointNet, mean error at severity 5 is 32.08%, below Source-only’s 35.88% and BFTT3D’s 33.78%; this closed-set result does not guarantee gains elsewhere.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.