MixerGANsformer: Can We Built A Strong GAN with Transformers and MLP-Mixers?
Official implementation of MixerGANsformer in PyTorch. A novel GAN model which consists of Transformers and MLP-Mixer. Preprint will be published soon.
Overview
In this model, generator is the the same structure in TransGAN's generator and the discriminator is from MLP-Mixer. The goal is to create an strong GAN model without convolutions and show that the MLP-Mixer and Transformers may help to create a strong GAN instead of using pure Transformers or MLP-Mixer in GANs.
Usage
Before running train.py, check whether you have libraries in requirements.txt! To save your model during training, create ./checkpoint folder using mkdir checkpoint.
Dataset
CIFAR10
Training
python train.py
License
MIT