This is a reimplementation from scratch of the VQ-GAN / transformer framework from "Taming Transofmers" to generate high resolution images
Find the link to donwload the dataset (kaggle)
Download packages pip install pytorch torchvision lpips panda
With the right CUDA version for you gpu
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!
Change the path & parameters in precompute_codebook.py
Run the program, it creates the dataset of Tensor for the Transformers
Change the path & parameter in train_stage2_transformer.py
Run the program to train the Transformers.
It checkpoins in checkpoints/
Change the model & path in samples.py
Run the models it create an images !