UDA-COPE: Unsupervised Domain Adaptation for Category-level Object Pose Estimation

Learning to estimate object pose often requires ground-truth (GT) labels,such as CAD model and absolute-scale object pose, which is expensive andlaborious to obtain in the real world. To tackle this problem, we propose anunsupervised domain adaptation (UDA) for category-level object pose estimation,called UDA-COPE. Inspired by recent multi-modal UDA techniques, the proposedmethod exploits a teacher-student self-supervised learning scheme to train apose estimation network without using target domain pose labels. We alsointroduce a bidirectional filtering method between the predicted normalizedobject coordinate space (NOCS) map and observed point cloud, to not only makeour teacher network more robust to the target domain but also to provide morereliable pseudo labels for the student network training. Extensive experimentalresults demonstrate the effectiveness of our proposed method bothquantitatively and qualitatively. Notably, without leveraging target-domain GTlabels, our proposed method achieved comparable or sometimes superiorperformance to existing methods that depend on the GT labels.