This study proposes a novel approach that improves upon existing object pose estimation methods, which only consider the front surface of an object. This approach predicts the 3D coordinates of the front, back, and internal surfaces of an object, and generates ultra-high-density 2D-3D correspondences. Specifically, Hierarchical Continuous Coordinate Encoding (HCCE) is utilized to represent the coordinates of the front and back surfaces more accurately and efficiently. The proposed method outperforms existing state-of-the-art approaches on the Bopp-Op dataset.
