Learning a Disentangled Embedding for Monocular 3D Shape Retrieval and Pose Estimation (1812.09899v2)

Abstract: We propose a novel approach to jointly perform 3D shape retrieval and pose estimation from monocular images.In order to make the method robust to real-world image variations, e.g. complex textures and backgrounds, we learn an embedding space from 3D data that only includes the relevant information, namely the shape and pose. Our approach explicitly disentangles a shape vector and a pose vector, which alleviates both pose bias for 3D shape retrieval and categorical bias for pose estimation. We then train a CNN to map the images to this embedding space, and then retrieve the closest 3D shape from the database and estimate the 6D pose of the object. Our method achieves 10.3 median error for pose estimation and 0.592 top-1-accuracy for category agnostic 3D object retrieval on the Pascal3D+ dataset, outperforming the previous state-of-the-art methods on both tasks.