Towards Faster and Stabilized GAN Training for High-fidelity Few-shot Image Synthesis

TL;DR

This paper introduces a lightweight GAN that achieves high-quality, high-resolution image synthesis with minimal training data and computational resources, enabling fast and stable training from scratch.

Contribution

The authors propose a novel lightweight GAN architecture with specific modules, enabling high-fidelity image synthesis with limited data and training time, outperforming existing methods under constrained resources.

Findings

Achieves 1024x1024 resolution with few hours of training on a single GPU.

Performs well with less than 100 training samples across diverse datasets.

Outperforms StyleGAN2 in limited-data and limited-resource scenarios.

Abstract

Training Generative Adversarial Networks (GAN) on high-fidelity images usually requires large-scale GPU-clusters and a vast number of training images. In this paper, we study the few-shot image synthesis task for GAN with minimum computing cost. We propose a light-weight GAN structure that gains superior quality on 1024*1024 resolution. Notably, the model converges from scratch with just a few hours of training on a single RTX-2080 GPU, and has a consistent performance, even with less than 100 training samples. Two technique designs constitute our work, a skip-layer channel-wise excitation module and a self-supervised discriminator trained as a feature-encoder. With thirteen datasets covering a wide variety of image domains (The datasets and code are available at: https://github.com/odegeasslbc/FastGAN-pytorch), we show our model's superior performance compared to the state-of-the-art StyleGAN2, when data and computing budget are limited.