Unsupervised Learning of 3D Structure from Images
This work addresses the challenge of unsupervised 3D reconstruction in computer vision, which is foundational for applications in robotics and AR/VR.
The paper tackles the problem of recovering 3D structure from 2D images by learning deep generative models and performing probabilistic inference, achieving high-quality samples and establishing the first benchmarks on datasets like ShapeNet.
A key goal of computer vision is to recover the underlying 3D structure from 2D observations of the world. In this paper we learn strong deep generative models of 3D structures, and recover these structures from 3D and 2D images via probabilistic inference. We demonstrate high-quality samples and report log-likelihoods on several datasets, including ShapeNet [2], and establish the first benchmarks in the literature. We also show how these models and their inference networks can be trained end-to-end from 2D images. This demonstrates for the first time the feasibility of learning to infer 3D representations of the world in a purely unsupervised manner.