Source code for the paper "Invisible Black-Box Backdoor Attack against Deep Cross-Modal Hashing Retrieval".
- python == 3.7.10
- pytorch == 1.4.0
- torchvision == 0.2.1
- numpy == 1.19.2
- h5py == 3.4.0
- scipy == 1.7.1
We use three cross-modal datasets for experiments. Since MS-COCO do not have common text features, we use the pre-trained BERT model to extract 1024-dimension text features. All datasets are available by the following link:
- FLICKR-25K: https://pan.baidu.com/s/1Ie9PDqC9mAmBdxqX0KJ0ng
Password: yjkd - MS-COCO: https://pan.baidu.com/s/1ocZTVx1GFFdceoSYbIWkbQ
Password: 2a6l - NUS-WIDE: https://pan.baidu.com/s/1Yvqt4Bdjsq1gPaJn2IqIEw
Password: doi1
We provide an knockoff of 32-bit DCMH on the FLICKR-25K dataset. The knockoff can be obtained by the following link:
- The Trained 32-bit DCMH on FLICKR-25K: https://pan.baidu.com/s/1JcQd_SepWVz-Js4X8yqjPQ
Password: b6sd
We carry out backdoor attack for three cross-modal hashing methods, including DCMH, CPAH, DADH. All attacked hashing models can be obtained by the following link:
- Deep Cross-Modal Hashing (DCMH): https://github.com/WendellGul/DCMH
- Consistency-Preserving Adversarial Hashing (CPAH): https://github.com/comrados/cpah
- Deep Adversarial Discrete Hashing (DADH): https://github.com/Zjut-MultimediaPlus/DADH
Coming soon...