certainly cost intensive

[tanlingp]

Thanks for your excellent work.
I found that it took six hours to train just 1,000 images. This is certainly cost intensive. I would like to ask if this is a personal factor for me or for that model, and also would like to ask if that brute requirement of 1000 images? Would it be a smaller amount of data?
I look forward to your reply, thank you very much.

[WindVChen]

Hi @tanlingp ,

It does take time. Slow running speed is probably a common problem of most diffusion models at present. For a quick look at the results, maybe you can just run it on the demo images we provide.

[tanlingp]

At the risk of asking, why did I find a precision of 0.0% for both clean and adversarial samples in my test?

[WindVChen]

Sometimes it may be because the given input image itself is a difficult sample for the classifier. In this case, both the original clean image and the attacked image will lead to 0.0% accuracy.

[tanlingp]

Have you used imagenet-compatible generated adversarial samples for testing? It's incredible that his clean samples also have an accuracy of 0. Looking forward to your reply

[tanlingp]

The above problems have been solved and are mainly a matter of image sequencing. Thank you very much for your patience in answering. Thank you so much.

[WindVChen]

Glad to see the problem solved 👍 .