# A Three-Player GAN: Generating Hard Samples To Improve Classification Networks

## TL;DR

This paper introduces a three-player GAN framework where the generator creates challenging, realistic samples to enhance the robustness and accuracy of classification networks, demonstrated on traffic sign recognition.

## Contribution

It presents a novel three-player GAN architecture that synthesizes hard samples to improve classifier robustness without assuming specific augmentation types.

## Key findings

- Generator produces realistic, hard-to-classify samples.
- Classifier becomes more robust after training with generated samples.
- Method improves classification accuracy on traffic sign dataset.

## Abstract

We propose a Three-Player Generative Adversarial Network to improve classification networks. In addition to the game played between the discriminator and generator, a competition is introduced between the generator and the classifier. The generator's objective is to synthesize samples that are both realistic and hard to label for the classifier. Even though we make no assumptions on the type of augmentations to learn, we find that the model is able to synthesize realistically looking examples that are hard for the classification model. Furthermore, the classifier becomes more robust when trained on these difficult samples. The method is evaluated on a public dataset for traffic sign recognition.