Hyperparticle / one-pixel-attack-keras

Keras implementation of "One pixel attack for fooling deep neural networks" using differential evolution on Cifar10 and ImageNet
https://arxiv.org/abs/1710.08864
MIT License
1.21k stars 213 forks source link

How does the targeted attack work? #20

Closed TudouJack closed 4 years ago

TudouJack commented 4 years ago

Hello: I feel puzzled about that if just changing one pixel can implement the targeted attack. I think only one-pixel change can not make the target that attacked with clear. Please tell me how does the targeted attack work with so few pixel. Thank you!

Hyperparticle commented 4 years ago

Targeted attack is much more difficult than untargeted, so it might be the case that you won't easily find a pixel in the image that will change the probability to the class of choice. However, it may be surprising that sometimes the attack succeeds. It's hard to know exactly why it's the case, it's still an active area of research.