Hi! I have implemented your results of defense in MNIST dataset, but the results is not as good as yours.
I have trained the defense model and the classify model, and download the data which you upload to the Baiduyun , and test the defense performance . My classify model accuracy is 99%+, and the mean_squared_error of two autoencoder model is less than 0.003. Finally,I got the following graph.
I think there may be a problem with the detector，but I dont known the reason!
Hi! I have implemented your results of defense in MNIST dataset, but the results is not as good as yours. I have trained the defense model and the classify model, and download the data which you upload to the Baiduyun , and test the defense performance . My classify model accuracy is 99%+, and the mean_squared_error of two autoencoder model is less than 0.003. Finally,I got the following graph.
I think there may be a problem with the detector，but I dont known the reason!