/WDAD

Adversarial sample detection based on weak dark textures

Primary LanguageJupyter NotebookMIT LicenseMIT

Adversarial sample detection based on weak dark textures

Background

Based on the 42nd PRP scientific research program, this project realizes the function of detecting whether images are antagonistic samples according to their gray-scale texture features.

Usage

Run ./PRP_demo/gui/ui.py to invoke the graphical interface.

Click the button at the top of the interface to select the picture, and the system will automatically detect whether it is a counter sample.

Related Efforts

The attack section is based on the demo of adversarial-attacks-pytorch.

The network is based on resnet50 net.

Maintainer

@faiimea

Contributors

@cfg554

TODO

  • Optimize code structure
  • Optimize GUI interface

License

MIT