UniPose9D: Universal Category-Agnostic Object Pose Estimation
UniPose9D is a universal model that estimates an object's full 9D pose (rotation, translation, and size) without needing category labels or reference models. It achieves this by combining visual features from DINOv2 with geometric features from a PointNet. The model predicts shape-aware coordinates using point pairs and uses an adaptive N-hop Kabsch–Umeyama algorithm to recover the final pose accurately.