Inverse Graphics GAN: Learning to Generate 3D Shapes from Unstructured 2D Data

TL;DR

This paper presents a scalable method for training 3D shape generative models from 2D images using an off-the-shelf renderer and a proxy neural renderer to handle non-differentiability, outperforming previous models.

Contribution

Introduces a novel training technique that leverages existing non-differentiable renderers with a proxy neural renderer and discriminator matching for 3D shape generation from 2D data.

Findings

Model generates higher quality 3D shapes than prior methods.

Effective training with only unstructured 2D images.

Utilizes standard industrial renderers without custom modifications.

Abstract

Recent work has shown the ability to learn generative models for 3D shapes from only unstructured 2D images. However, training such models requires differentiating through the rasterization step of the rendering process, therefore past work has focused on developing bespoke rendering models which smooth over this non-differentiable process in various ways. Such models are thus unable to take advantage of the photo-realistic, fully featured, industrial renderers built by the gaming and graphics industry. In this paper we introduce the first scalable training technique for 3D generative models from 2D data which utilizes an off-the-shelf non-differentiable renderer. To account for the non-differentiability, we introduce a proxy neural renderer to match the output of the non-differentiable renderer. We further propose discriminator output matching to ensure that the neural renderer learns to smooth over the rasterization appropriately. We evaluate our model on images rendered from our generated 3D shapes, and show that our model can consistently learn to generate better shapes than existing models when trained with exclusively unstructured 2D images.