PyTorch Implementation of AnimeGANv2

AnimeGAN2 - PyTorch

• This codes are forked from bryandlee's animegan2 -pytorch

Trained Result (25,000 steps, with 16batchs)

Update

• 2023-12-17 Added training codes
• 2024-01-04 Implemented image logging, save checkpoint
• 2024-01-07 Trained for cartoon dataset

Training

Step 1.

Make edge smoothing data

python utils/edge_smooth.py --root_dir {ANIME_IMAGE_DIRECTORY} --out_dir {EDGE_SMOOTH_SAVE_DIRECTORY} --image_size {EDGE_SMOOTH_IMAGE_SIZE}

Step 2.

Train

python train.py --photo_dir {PHOTO_IMAGE_DIRECTORY} --anime_dir {ANIME_IMAGE_DIRECTORY} --smooth_dir {SMOOTH_IMAGE_DIRECTORY}

Requirements

• Source repository's requirements
• Pytorch lightning