yuyi-sd / robust_rain_removal Goto Github PK

How to test the robustness of six rain removal networks with the adversarial attack? Should I use the code of these networks(JORDER-E, RCDNet...) for testing? I didn't find the code for these networks in the file. Hope you can guide me🙏.

Hi, how did you set the λ in the adversarial loss during training

Follow your guide， I can run the code， but i cannot get the result properly，PSNR alwyas nan.
So，can you refresh your guide/code or teach me how to use your code?
Thank you very much

Hi, thanks for your work, and I have a problem with the LMSE attack and the LPIPS attack proposed in your paper. Why does LPIPS attack targets machine vision, and LMSE attack targets human vision？

Hi @yuyi-sd , I have read your paper carefully and I‘m very interested in this part.
And I would appreciate it if you could share your code for training.
Looking forward to your reply~

Hi, I got nan for evaluation. Any idea about this? Thx~

python robust.py --data_test RainHeavyTest --ext img --pre_train ../experiments/MPRNet_R_SEADD_MB_robust_pgd_Rain100H_e4/model/model_latest.pt --model MPRNet_R_SEADD_MB --test_only --save_results --save_gt --save_attack --save MPRNet_R_SEADD_MB_robust_pgd_test_Rain100H_e4_1 --n_GPUs 1 --attack_iters 20 --robust_epsilon 1 --robust_alpha 0.25

../data/test/small/
Making model...
rest
Loading model from ../experiments/MPRNet_R_SEADD_MB_robust_pgd_Rain100H_e4/model/model_latest.pt

Evaluation:
0it [00:00, ?it/s]
[RainHeavyTest x2] PSNR: nan (Best: nan @epoch 1)
[RainHeavyTest x2] sr PSNR: nan (Best: nan @epoch 1)
[RainHeavyTest x2] lr PSNR: nan (Best: nan @epoch 1)
Total time: 0.03s