Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs

TL;DR

This paper introduces FreezeD, a simple yet effective fine-tuning method for GANs that involves freezing lower discriminator layers, outperforming previous transfer learning techniques across various datasets and architectures.

Contribution

The paper proposes FreezeD, a straightforward baseline for fine-tuning GANs by freezing lower discriminator layers, which improves performance over existing methods.

Findings

FreezeD outperforms previous transfer learning techniques in GAN fine-tuning.

The method is effective across multiple datasets and GAN architectures.

Freezing lower discriminator layers is a surprisingly strong baseline.

Abstract

Generative adversarial networks (GANs) have shown outstanding performance on a wide range of problems in computer vision, graphics, and machine learning, but often require numerous training data and heavy computational resources. To tackle this issue, several methods introduce a transfer learning technique in GAN training. They, however, are either prone to overfitting or limited to learning small distribution shifts. In this paper, we show that simple fine-tuning of GANs with frozen lower layers of the discriminator performs surprisingly well. This simple baseline, FreezeD, significantly outperforms previous techniques used in both unconditional and conditional GANs. We demonstrate the consistent effect using StyleGAN and SNGAN-projection architectures on several datasets of Animal Face, Anime Face, Oxford Flower, CUB-200-2011, and Caltech-256 datasets. The code and results are available at https://github.com/sangwoomo/FreezeD.