Questions about the training process

[FrankHong77]

The training code is not provided in the GitHub repository, I would like to know:

1. Obviously the inference process uses only style blocks. But is the training process for all blocks, or just style blocks?（I tried both method and found that training only style block didn't work well）
2. Does the training process require stylized images as training target? If so, how do you get stylized images?