Refinement Based on Contour Alignment for Monocular 6D Object Pose Estimation

In this paper, we present a novel refinement technique for object pose estimation. We perform the extraction of the Contour of Interest (CoI) from both the visible mask and the complete mask. Following this, we utilize both the Col and the contour rendered by the initial pose to optimize the pose. Our fundamental insight is that contour alignment serves as an essential criterion for accurately estimating the pose, while also facilitating the generation of gradients for pose optimization. Notably, our method diverges from the render-and-compare approach by solely requiring the rendering of the segmentation, without the need for training a new refinement network. This distinctive aspect eliminates the time-consuming nature associated with the refinement process. To validate the effectiveness of our proposed technique, we conduct a comprehensive evaluation on benchmark datasets, namely YCB-V and IC-BIN. The code can be found at https://github.com/lyltc1/ContourAlignment.