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

Sangwoo Mo, Minsu Cho, Jinwoo Shin
2/25/2020
cs.CVcs.LGstat.ML

Official code: sangwoomo/FreezeD (286 stars, last commit 2021-12-12).

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.

View on arXivView PDF

Code Implementations(1)

sangwoomo/FreezeDOfficial100%

Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs (CVPRW 2020)

28635Shell, PythonFeb 25, 20204 years ago
gangenerative-adversarial-networkgenerative-modelstransfer-learning

Similar Papers

Differentiable Augmentation for Data-Efficient GAN Training

86% match1 repo1.3k2020

Training Generative Adversarial Networks with Limited Data

86% match1 repo1.8k2020

SAN: Inducing Metrizability of GAN with Discriminative Normalized Linear Layer

86% match1 repo982023

Benign Overfitting in Adversarial Training for Vision Transformers

85% match2026

Bridging Optimal Transport and Jacobian Regularization by Optimal Trajectory for Enhanced Adversarial Defense

85% match2023

DADA: Deep Adversarial Data Augmentation for Extremely Low Data Regime Classification

85% match1 repo522018

Cite this paper

@article{mo2020freeze,
  title  = {Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs},
  author = {Sangwoo Mo and Minsu Cho and Jinwoo Shin},
  year   = {2020},
  eprint = {2002.10964},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2002.10964}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2002.10964.svg)](https://codeofpaper.com/paper/2002.10964)

Discussion