AnimeGAN: A Novel Lightweight GAN for Photo Animation

Unofficial pytorch-lightning implementation of a paper, "AnimeGAN: A Novel Lightweight GAN for Photo Animation"

• original tensorflow official implementation is here v1 and v2
• pytorch implementation is here

Requirements

need for training

• torch==1.10.1
• torchvision==0.11.2
• pytorch-lightning==1.5.7
• tqdm==4.62.3
• wandb==0.12.10
• opencv-python-headless==4.5.5.62
• albumentations==1.1.1
• easydict==1.9

need for test and inference

• onnxruntime==1.10.0
• moviepy==1.0.3
• pytest==7.0.0
• pytest-watch==4.2.0
• pytest-testmon==1.2.3
• pytest-cov==3.0.0

Install requirements by pip

pip install -r requirements.txt

Install requirements by docker

docker build .

Pretrained weights / models download

1. Shinkai style training

• checkpoint
• torchscript
• onnx

DATASET

• dataset download from original AnimeGAN repo "download"

• download and unzip dataset

wget -O anime-gan.zip https://github.com/ptran1203/pytorch-animeGAN/releases/download/v1.0/dataset_v1.zip
unzip anime-gan.zip

project root
├── anime-gan.zip
├── dataset
│   ├── Hayao       # --style_image_root
│   │   ├── smooth
│   │   └── style
│   ├── Kimetsu     # --style_image_root
│   │   ├── smooth
│   │   └── style
│   ├── Paprika     # --style_image_root
│   │   ├── smooth
│   │   └── style
│   ├── Shinkai     # --style_image_root
│   │   ├── smooth
│   │   └── style
│   ├── SummerWar
│   ├── test
│   └── train_photo # --real_image_root
├── main.py
...

Training

1. prepare dataset

2. start training

python main.py \
--seed 2022 \
--real_image_root "./dataset/train_photo" \
--style_image_root "./dataset/Shinkai" \
--device "cuda:0" \
--image_channels 3 \
--image_size 256 \
--batch_size 16 \
--num_workers 8 \
--g_dim 32 \
--d_dim 32 \
--d_layers 3 \
--sn "True" \
--epochs 300 \
--init_epochs 10 \
--init_lr 0.0002 \
--g_lr 0.00002 \
--d_lr 0.00004 \
--beta_1 0.5 \
--beta_2 0.999 \
--ctlw 1.5 \
--stlw 3.0 \
--colw 30.0 \
--advgw 10.0 \
--advdw 10.0 \
--show_image_count 8

Training result

Shinkai

Input
Output

Input	Output

References

• AnimeGAN
• pytorch-animeGAN