Inquiry about Running on My Dataset

[KeepNoob]

If I want to run DiffusionGuard on my own dataset, I have to generate a set of masks of my data and then run the main.py? If so, which method do you use to generate a mask?

[choi403]

Hi, thank you so much for your interest in our work.
If you want to run DiffusionGuard on your own dataset, you will have to prepare two images: the original image to be protected, and the mask to specify which region you would like to protect.
The two images need to be of the same dimension (width and height).

For the mask image, yes, you need to generate your own mask.
Although there are numerous ways to produce a mask such as using the SAM (Segment Anything) model, an easy method is to use a painting tool such as Photoshop or MS Paint.
You can simply set the background to black, and use a white circular brush to indicate the region which you would like to protect.
Please refer to the example mask in our github page.