Skip to content

LTeuse/Report-Taming-Transformers

Repository files navigation

Report-Taming-Transformers

This is a reimplementation from scratch of the VQ-GAN / transformer framework from "Taming Transofmers" to generate high resolution images

Run & Reproduce Results

0: Download Animal Faces (AFHQ) dataset

Find the link to donwload the dataset (kaggle) Download packages pip install pytorch torchvision lpips panda With the right CUDA version for you gpu

1: Train VQGAN

Change the parameter in train_stage1_vqgan.py with the right hyperparameters for the VQGAN and the right path for datasets, loging and checkpoints. Run the program, it trains and checkpoints in checkopints!

2: Convert dataset

Change the path & parameters in precompute_codebook.py Run the program, it creates the dataset of Tensor for the Transformers

3: Train Transformers

Change the path & parameter in train_stage2_transformer.py Run the program to train the Transformers. It checkpoins in checkpoints/

4: Samples new images

Change the model & path in samples.py Run the models it create an images !

About

This is a reimplementation from scratch of the VQ-GAN / transformer framework from "Taming Transofmers" to generate high resolution images

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages