Image GANs meet Differentiable Rendering for Inverse Graphics and Interpretable 3D Neural Rendering

TL;DR

This paper introduces a method to extract interpretable 3D information from GANs using differentiable rendering, enabling explicit 3D reasoning and controllable neural rendering, surpassing existing inverse graphics approaches.

Contribution

The authors propose a novel framework that leverages GANs as multi-view data generators and uses differentiable renderers to disentangle 3D properties, improving interpretability and performance.

Findings

Outperforms state-of-the-art inverse graphics networks

Enables controllable 3D neural rendering

Achieves significant quantitative and qualitative improvements

Abstract

Differentiable rendering has paved the way to training neural networks to perform "inverse graphics" tasks such as predicting 3D geometry from monocular photographs. To train high performing models, most of the current approaches rely on multi-view imagery which are not readily available in practice. Recent Generative Adversarial Networks (GANs) that synthesize images, in contrast, seem to acquire 3D knowledge implicitly during training: object viewpoints can be manipulated by simply manipulating the latent codes. However, these latent codes often lack further physical interpretation and thus GANs cannot easily be inverted to perform explicit 3D reasoning. In this paper, we aim to extract and disentangle 3D knowledge learned by generative models by utilizing differentiable renderers. Key to our approach is to exploit GANs as a multi-view data generator to train an inverse graphics network using an off-the-shelf differentiable renderer, and the trained inverse graphics network as a teacher to disentangle the GAN's latent code into interpretable 3D properties. The entire architecture is trained iteratively using cycle consistency losses. We show that our approach significantly outperforms state-of-the-art inverse graphics networks trained on existing datasets, both quantitatively and via user studies. We further showcase the disentangled GAN as a controllable 3D "neural renderer", complementing traditional graphics renderers.