# Feature Quantization Improves GAN Training

arXiv: https://arxiv.org/abs/2004.02088 · HTML: https://mlsanity.com/paper/2004.02088/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 5 Apr 2020.

## Built on by (most cited first, 23 of 23)

- [Improved Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2102.09672/) (2021)
- [Diff-Instruct: A Universal Approach for Transferring Knowledge From Pre-trained Diffusion Models](https://mlsanity.com/paper/2305.18455/) (2023)
- [One-Step Diffusion Distillation through Score Implicit Matching](https://mlsanity.com/paper/2410.16794/) (2024)
- [Training GANs with Stronger Augmentations via Contrastive Discriminator](https://mlsanity.com/paper/2103.09742/) (2021)
- [StudioGAN: A Taxonomy and Benchmark of GANs for Image Synthesis](https://mlsanity.com/paper/2206.09479/) (2022)
- [Styleformer: Transformer based Generative Adversarial Networks with Style Vector](https://mlsanity.com/paper/2106.07023/) (2021)
- [Flow Generator Matching](https://mlsanity.com/paper/2410.19310/) (2024)
- [Conditional GANs with Auxiliary Discriminative Classifier](https://mlsanity.com/paper/2107.10060/) (2021)
- [EGC: Image Generation and Classification via a Diffusion Energy-Based Model](https://mlsanity.com/paper/2304.02012/) (2023)
- [Manifold Learning Benefits GANs](https://mlsanity.com/paper/2112.12618/) (2021)
- [Expensive Multi-Objective Bayesian Optimization Based on Diffusion Models](https://mlsanity.com/paper/2405.08674/) (2024)
- [Self-Supervised GANs with Label Augmentation](https://mlsanity.com/paper/2106.08601/) (2021)
- [Artificial intelligence approaches for materials-by-design of energetic materials: state-of-the-art, challenges, and future directions](https://mlsanity.com/paper/2211.08179/) (2022)
- [Robust Vector Quantized-Variational Autoencoder](https://mlsanity.com/paper/2202.01987/) (2022)
- [Compress Guidance in Conditional Diffusion Sampling](https://mlsanity.com/paper/2408.11194/) (2024)
- [The Nuts and Bolts of Adopting Transformer in GANs](https://mlsanity.com/paper/2110.13107/) (2021)
- [CoopInit: Initializing Generative Adversarial Networks via Cooperative Learning](https://mlsanity.com/paper/2303.11649/) (2023)
- [StyleGenes: Discrete and Efficient Latent Distributions for GANs](https://mlsanity.com/paper/2305.00599/) (2023)
- [Scalable Balanced Training of Conditional Generative Adversarial Neural Networks on Image Data](https://mlsanity.com/paper/2102.10485/) (2021)
- [A Unified Generative Adversarial Network Training via Self-Labeling and Self-Attention](https://mlsanity.com/paper/2106.09914/) (2021)
- [TL-GAN: Improving Traffic Light Recognition via Data Synthesis for Autonomous Driving](https://mlsanity.com/paper/2203.15006/) (2022)
- [Brain Tumor Synthetic Data Generation with Adaptive StyleGANs](https://mlsanity.com/paper/2212.01772/) (2022)
- [Enhancing Stability in Training Conditional Generative Adversarial Networks via Selective Data Matching](https://mlsanity.com/paper/2107.08792/) (2021)

## Abstract (arXiv)

The instability in GAN training has been a long-standing problem despite remarkable research efforts. We identify that instability issues stem from difficulties of performing feature matching with mini-batch statistics, due to a fragile balance between the fixed target distribution and the progressively generated distribution. In this work, we propose Feature Quantization (FQ) for the discriminator, to embed both true and fake data samples into a shared discrete space. The quantized values of FQ are constructed as an evolving dictionary, which is consistent with feature statistics of the recent distribution history. Hence, FQ implicitly enables robust feature matching in a compact space. Our method can be easily plugged into existing GAN models, with little computational overhead in training. We apply FQ to 3 representative GAN models on 9 benchmarks: BigGAN for image generation, StyleGAN for face synthesis, and U-GAT-IT for unsupervised image-to-image translation. Extensive experimental results show that the proposed FQ-GAN can improve the FID scores of baseline methods by a large margin on a variety of tasks, achieving new state-of-the-art performance.