Conclusion & Future work

Conclusion

In conclusion, our goal is this project is to transfer style in the Chinese font domain. We tested on well-know algorithm pix2pix but it does not perform well. After that, we implemented CycleGAN and DiscoGAN and got an acceptable performance. We tried to better the result so we modified CycleGAN and that resulted in a model that produce the best result in our case.

Future work

An useful extension of this work may be developing a method that can learn new style by using exiting model instead of having to train everything from scratch. Another interesting idea may be to test if it is possible to transfer style from one language to another.