search

bethgelab / foolbox

A Python toolbox to create adversarial examples that fool neural networks in PyTorch, TensorFlow, and JAX
https://foolbox.jonasrauber.de
MIT License
2.73k stars 422 forks source link

DeepFool cannot works normally on GluonCV models #432

Closed yqtianust closed 4 years ago

yqtianust commented 4 years ago

Hi,

We applied Foolbox DeepFool attack on GluonCV models. The 1000 input images are selected from ImageNet. However, we cannot get a reasonable success rate on several models: densenet121/169/201, mobilenet, resnet50/101/152.

For example: for model Desnet121, the success rate is only 0.0413971539 (32/773). And it takes quite a long time. Experiment Setting: untargeted attack, with distance foolbox.distances.MSE,

We can get reasonable success rates on vgg16/19 and alexnet.

If we switch DeepFool to FGSM and BIM, all the above models work well.

Thanks.

jonasrauber commented 4 years ago

Is this still relevant?