search

fra31 / auto-attack

Code relative to "Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks"
https://arxiv.org/abs/2003.01690
MIT License
639 stars 111 forks source link

Add FSGM_APR_SP #69

Closed iCGY96 closed 3 years ago

iCGY96 commented 3 years ago

Paper: Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain

Venue: ICCV 2021

Dataset and threat model: CIFAR-10, L-inf, 8/255

Code: code

Pre-trained model: model

Log file: log

Additional data: no

Clean and robust accuracy: 84.30% / 45.70%

Architecture: RN-18

Description of the model/defense: Motivated by the powerful generalizability of the human, we argue that reducing the dependence on the amplitude spectrum and enhancing the ability to capture phase spectrum can improve the robustness of CNN.

fra31 commented 3 years ago

Hi,

thanks for the submission! I'd add your model, with the evaluation on the full test set, to RobustBench since it is currently up-to-date, if this is fine for you.

iCGY96 commented 3 years ago

Very Thanks.

fra31 commented 3 years ago

Added with https://github.com/RobustBench/robustbench/commit/a97f10d6523017f547fb4a181b7933e54c8eb46b. If it looks fine to you, I update the website too.

iCGY96 commented 3 years ago

Thank you for your evaluation. RobustBench is a very good platform. We are not going to add the current results to RobustBench now. We will submit more results uniformly when we are ready for more models about cifar10-c and cifar100-c.