How to use diffpure to clean an already attacked adversarial image?

[pearlmary]

Hi,

How do I feed an input adversarial image for diffpure adversarial cleaning?

[weilinie]

You can directly pass the adversarial image to model.forward() in SDE_Adv_Model. By default, it will output a classification score and also save the purified image.

[pearlmary]

Hi, Thanks for the reply. Still I'm unable to get it. Can you guide me by listing out the step by step instructions on how to clean an adversarial image input.

[kabachuha]

@pearlmary I'm building a gradio extension (NV's license preserved) for Automatic1111's Stable Diffusion webui using the simple model inference call https://github.com/kabachuha/sd-webui-diffpure. It should allow for convenient processing of user selected images

[pearlmary]

Thanks. Let me check it out.

[zhongguadepi]

您好， 感谢您的回复。 我仍然无法得到它。 您能否列出有关如何清理对抗性图像输入的分步说明来指导我。

Hello, have you resolved how to provide input adversarial images for diffpure adversarial cleaning?