search

ZJZAC / Deep-Model-Watermarking

MIT License
66 stars 9 forks source link

about SR attack #5

Closed Jayamoon59 closed 11 months ago

Jayamoon59 commented 12 months ago

Hi, I've been learning this code recently. I want you to ask me some questions.

At present, I run./SR attack, which uses derain treatment. The main problem is that I don't know what data sets are included in the training input. What's the output?

There is also a small problem in running./Initial stage. I input a color image, but the output is four lines of images, all of which are black and white.

I was hoping you could help me out.

ZJZAC commented 11 months ago

For the SR attack, we utilize the code of official pix2pix, which learns a supervised image-to-image task. The data shall be a pair of rainy image A and the watermarked derained image B' for deraining tasks. If you would like more details on constructing the dataset, you can refer to the paper.

For initial training, the default setting is for debone task, therefore the output is a gray-scale image. You can easily change the pre-processing and the input channel for Embedding Net to adjust the deraining tasks.

Thanks for your interest in our work!