Face Generation using Discrete VAE

Please, follow the following steps for model training and evaluation.

python train_dvae.py --img_folder data/CelebAMask-HQ/CelebA-HQ-img --batch_size 64

python generate_dvae_embeddings.py --img_folder data/CelebAMask-HQ/CelebA-HQ-img --batch_size 128 --img_size 128

python train_transformer.py --learning_rate 0.0003 --batch_size 32  --array_folder data/dvae_codes/

Generate Images

python generate_images.py --vae_checkpoint_path checkpoints/vae_ckpt.pth --transformer_checkpoint_path checkpoints/transformer_ckpt.pt

The generated images are placed in results/transformer_training folder.

Name		Name	Last commit message	Last commit date
Latest commit History 18 Commits
checkpoints		checkpoints
data		data
results		results
.gitignore		.gitignore
README.md		README.md
dataset_dvae.py		dataset_dvae.py
dataset_transformer.py		dataset_transformer.py
generate_dvae_embeddings.py		generate_dvae_embeddings.py
generate_images.py		generate_images.py
model_dvae.py		model_dvae.py
model_mingpt.py		model_mingpt.py
model_transformer.py		model_transformer.py
test.sh		test.sh
train.sh		train.sh
train_dvae.py		train_dvae.py
train_transformer.py		train_transformer.py