HoloGAN: Unsupervised Learning of 3D Representations From Natural Images

TLDR

Most generative models rely on 2D kernels and make few assumptions about the 3D world, causing blurry images or artifacts in tasks that require strong 3D understanding such as novel‑view synthesis. The authors propose HoloGAN, a GAN that learns 3D representations from natural images and renders them realistically, enabling unsupervised 3D learning. HoloGAN learns explicit 3D features, offers rigid‑body pose control, and can be trained end‑to‑end from unlabelled 2D images without pose labels, 3D shapes, or multiple views. Experiments demonstrate that explicit 3D features allow HoloGAN to disentangle pose, identity, shape, and appearance while achieving visual quality comparable or superior to other models, making it the first generative model to learn 3D representations from natural images in an entirely unsupervised manner.

Abstract

We propose a novel generative adversarial network (GAN) for the task of unsupervised learning of 3D representations from natural images. Most generative models rely on 2D kernels to generate images and make few assumptions about the 3D world. These models therefore tend to create blurry images or artefacts in tasks that require a strong 3D understanding, such as novel-view synthesis. HoloGAN instead learns a 3D representation of the world, and to render this representation in a realistic manner. Unlike other GANs, HoloGAN provides explicit control over the pose of generated objects through rigid-body transformations of the learnt 3D features. Our experiments show that using explicit 3D features enables HoloGAN to disentangle 3D pose and identity, which is further decomposed into shape and appearance, while still being able to generate images with similar or higher visual quality than other generative models. HoloGAN can be trained end-to-end from unlabelled 2D images only. Particularly, we do not require pose labels, 3D shapes, or multiple views of the same objects. This shows that HoloGAN is the first generative model that learns 3D representations from natural images in an entirely unsupervised manner.