Add to Chrome
Add to Firefox
Add to Edge3D Model-based Zero-Shot Pose Estimation Pipeline
May 29, 2023
Most existing learning-based pose estimation methods are typically developed for non-zero-shot scenarios, where they can only estimate the poses of objects present in the training dataset. This setting restricts their applicability to unseen objects in the training phase. In this paper, we introduce a fully zero-shot pose estimation pipeline that leverages the 3D models of objects as clues. Specifically, we design a two-step pipeline consisting of 3D model-based zero-shot instance segmentation and a zero-shot pose estimator. For the first step, there is a novel way to perform zero-shot instance segmentation based on the 3D models instead of text descriptions, which can handle complex properties of unseen objects. For the second step, we utilize a hierarchical geometric structure matching mechanism to perform zero-shot pose estimation which is 10 times faster than the current render-based method. Extensive experimental results on the seven core datasets on the BOP challenge show that the proposed method outperforms the zero-shot state-of-the-art method with higher speed and lower computation cost.
