A PyTorch Implementation of "Evade Deep Image Retrieval by Stashing Private Images in the Hash Space" accepted by CVPR 2020

It is based on HashNet pytorch version, tested on PyTorch 0.3.1.

Python 2.7

PyTorch 0.3.1 (Newer version may work with a little modification but we cannot guarantee.)

A CUDA device

ImageNet, Places365, CIFAR-10, Fashion-MNIST and MNIST are supported.

We use a subset that includes 10% of all classes of the ImageNet following HashNet's implementation, which contains only 100 class. You may download them via thuml's link.

Remember to modify the path in files in ./HashNet/pytorch/data/imagenet/ directory.

We use a subset that includes 10% of all classes of Places365. Please use this link to download the whole set.

Relative path is used. Please check ./HashNet/pytorch/data/places365_standard/ directory for where to place the train and the val folders.

We use the full dataset of these tiny datasets. Since they are all supported by torchvision package, there is no need to download the original dataset.

Running our algorithm can be easy to do by simply typing:

python myGetAdv.py 

However, before typing it, a model file and hash code files for corresponding dataset should be ready to make it run smoothly. The model file should be saved into ./HashNet/pytorch/snapshot/[job_dataset]/48bit/[net]_hashnet/. The hash code files should be saved into ./HashNet/pytorch/src/save_for_load/blackbox/[net]/[job_dataset]/.

We provide link for downloading all models and hash code files we have used. Here is the link. (This method is RECOMMENDED)

The trained models are created by myExtractCodeLabel.py after the dataset are ready. Run it using python myTrain.py

The Hash Code files are created from myExtractCodeLabel.py after setting up the trained models. Run it using python myExtractCodeLabel.py.

Both of them support all the five datasets we mentioned earlier. For the details, you may need to dig into them.

Yanru Xiao: [[email protected]](mailto:[email protected])
For any further issues, please oepn a new issue on github.