search

kai-wen-yang / CD-VAE

[NeurIPS 2021] "Class-Disentanglement and Applications in Adversarial Detection and Defense"
MIT License
43 stars 8 forks source link

cifar, the acc of training set is lower than that of test set? #3

Closed chuchuhe closed 2 years ago

chuchuhe commented 2 years ago

I ran the original code 'distentangle_cifar.py' and got the below result. Is it the same result as yours? Is it correct? If so, I am confused that why the acc of training set is lower than that of test set. 微信图片_20220310194956

kai-wen-yang commented 2 years ago

I ran the original code 'distentangle_cifar.py' and got the below result. Is it the same result as yours? Is it correct? If so, I am confused that why the acc of training set is lower than that of test set. 微信图片_20220310194956

Training uses random augmentation which is difficult to classify, testing does not use data augmentation.