<<<<<<< HEAD <<<<<<< HEAD
=======
4f9a70bfec15a9e143ff83d1f657a2558121f534 =======
4f9a70bfec15a9e143ff83d1f657a2558121f534 I have changed some source codes so that Feature-dis can be tested on dataset Cifar10.
To run this code, you have to follow these three steps:
-
- make a new directory named checkpoint/cifar10, and put a pre-trained network model file (wide-resnet-28x10.t7) in this new created directory.
-
- make a new directory named data, and put data files of Cifar10 here.
-
- run demo.sh
This repository implements DNN-Oriented JPEG compression introduced in the following paper
Feature Distillation: DNN-Oriented JPEG Compression Against Adversarial Examples