Closed lzyhha closed 3 years ago
Hi @lzyhha, I downloaded pre-trained model of SEAM at the official SEAM repository (https://github.com/YudeWang/SEAM), and applied adversarial climbing similar to obtain_CAM_masking.py
.
We observed that the values of the optimal hyper-parameters differ according to the backbone network, especially $T$ and $\lambda$. SEAM and SC-CAM used Wide ResNet-38. I'm not sure, but I think $T$ was 8 and $\lambda$ was 2.
Thanks! Your reply is very helpful.
Hi @lzyhha ! I notice that you can reproduce the result of "adv_cam+SEAM" by yourself. Can you share the code with me? Thanks a lot!
Hello, thanks for your great work. I notice that you have combine the AdvCAM with SEAM in paper. Could you please teel me how ti combine them?