hoang1007 / face2anime

Anime Face Style Transfer

Anime Style Transfer

This is the final project of the Computer Vision course at Vingroup AI Engineer Training Program 2023. In this project, we propose a novel method for style transfer between anime and real images using Cycle GAN.

Setup

To setup the project, run the following command:

pip install -v -e .

Data preparation

Download datasets from the following links and put them in the data folder:

• Shinkai Makoto's scenery dataset

Note: For other datasets described in the report, you can download out preprocessed data from here

Training

To train the model, run the following command:

python train.py --config CONFIG_FILE_PATH

where CONFIG_FILE_PATH is the path to the config file. For example, to train the model with the default config, run:

python train.py --config configs/default.yaml

when the training is done, the model will be saved in the checkpoints folder.

Inference

To generate images using the trained model, run the following command:

python scripts/inference.py --checkpoint CHECKPOINT_PATH --image_path IMAGE_PATH

where CHECKPOINT_PATH is the path to the trained model, IMAGE_PATH is the path to the input image. For example, to generate an image using the default config, run:

python scripts/inference.py --checkpoint checkpoints/lastest.ckpt --image_path data/landscape/landscape.jpg

The generated image will be saved as output.jpg in the current directory.

Experiments

You can track our experiments via Weights & Biases 🥰.

Using the pre-trained models

You can download our pre-trained models with following links:

• Anime face style transfer
• Anime scenery style transfer