Some issues

[zhangsong1213]

Hello, thank you very much for sharing your code. However, I have noticed that some parts of the code do not align with the content of the paper.

1. The code following PAdaIN does not match the paper content.
2. I couldn't find the preprocessing method for generating reference images in the code.
3. The balance module and stretch module involve convolution, and the convolution parameters need to be trained before use.
   Additionally, the essence of the paper is essentially based on supervised training using reference images, which is quite different from the title of the paper.

[alzayats]

Hello,

Thank you for your interest in our work and for taking the time to review our code. We appreciate your feedback and understand your concerns. Let me address them one by one:

1. PAdaIN Code: The implementation of PAdaIN can be found in the file UDnet.py on lines 102-109. We believe this section of the code aligns with the content of the paper.

2. Preprocessing Method: The preprocessing method for generating reference images is implemented in the file label_gen.py.

3. Balance and Stretch Modules: The balance and stretch modules are trained online. This process is implemented on line 77 in the file UDnet.py.

4. Supervised Training: The paper relies on auto-generated reference images without ground truth, see dataset.py lines 71-115.

We hope this clarifies your concerns. Please feel free to reach out if you have any more questions or suggestions.