Instance-Conditioned GAN

TL;DR

This paper introduces IC-GAN, a novel instance-conditioned GAN that models complex data distributions by learning around individual data points, leading to improved image generation on datasets like ImageNet and COCO-Stuff.

Contribution

The paper presents a non-parametric, instance-conditioned approach to GANs that enhances modeling of complex datasets and enables flexible transfer to unseen datasets.

Findings

IC-GAN outperforms unconditional models on ImageNet and COCO-Stuff.

IC-GAN can transfer to new datasets by changing conditioning instances.

IC-GAN achieves competitive results on ImageNet with controllable generation.

Abstract

Generative Adversarial Networks (GANs) can generate near photo realistic images in narrow domains such as human faces. Yet, modeling complex distributions of datasets such as ImageNet and COCO-Stuff remains challenging in unconditional settings. In this paper, we take inspiration from kernel density estimation techniques and introduce a non-parametric approach to modeling distributions of complex datasets. We partition the data manifold into a mixture of overlapping neighborhoods described by a datapoint and its nearest neighbors, and introduce a model, called instance-conditioned GAN (IC-GAN), which learns the distribution around each datapoint. Experimental results on ImageNet and COCO-Stuff show that IC-GAN significantly improves over unconditional models and unsupervised data partitioning baselines. Moreover, we show that IC-GAN can effortlessly transfer to datasets not seen during training by simply changing the conditioning instances, and still generate realistic images. Finally, we extend IC-GAN to the class-conditional case and show semantically controllable generation and competitive quantitative results on ImageNet; while improving over BigGAN on ImageNet-LT. Code and trained models to reproduce the reported results are available at https://github.com/facebookresearch/ic_gan.