This repository implements the DAG attack proposed by Xie et al. DAG is an adversarial attack for semantic segmentation DNNs. The attack generates an adversarial image against a target which closely resembles the real image while fooling a state of the art segmentation DNN.
- PyTorch > 0.4.0
- Numpy
- Matplotlib
- Random
- CUDA
The Jupyter Notebook contains an example of attacking a UNet network trained on the Oasis dataset.
Paschali, M., Conjeti, S., Navarro, F., & Navab, N. (2018, September). Generalizability vs. robustness: investigating medical imaging networks using adversarial examples. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 493-501). Springer, Cham.
Xie, C., Wang, J., Zhang, Z., Zhou, Y., Xie, L., & Yuille, A. (2017). Adversarial examples for semantic segmentation and object detection. In Proceedings of the IEEE International Conference on Computer Vision (pp. 1369-1378).