code for larger network
Opened this issue · 2 comments
yulu0724 commented
Hi, thanks for implementing the ZOO paper in pytorch, it is quite useful.
I am wondering if your code can be also used for larger newtork (e.g. Inception-v3). As shown in the paper, there are some tricks for this large network.
Thanks!
as791 commented
Hi, I am glad you find the implementation useful.
Although, the code is not yet fully compatible for larger image size data like ImageNet (I think you want to use inception v3 for dataset like that) as it requires the functionality of importance sampling, hierarchical attack, and dimentional reduction.
I have yet to include them in the code and they will be included soon.
Let me know if any other help needed.