This is a application demo of CycleGAN, filling colors automatically based on black-and-white line drafts of a fixed size of 256x256.
To run the code, an environment of TensorFlow and Keras is needed.
To train the model, you have to prepare 4 datasets:
- trainA: Line drafts for training.
- trainB: Colored avaters for training.
- testA: Line drafts for testing.
- testB: Colored avaters for testing.
To start training, open cmd and input:
python main.py -t
Training options can be modified in main.py.
Codes are edited from gan-zoo on GitHub.
First row: line draft to colored image
Second row: colored image to line draft
And here are some training results during the training period.
Original sketch:
Results during the training:
