Skip to content
ROSA: Robust Salient Object Detection against Adversarial Attacks, IEEE Transactions on Cybernetics, 2019
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
LICENSE
README.md
crf.prototxt
dss_train.prototxt
mscaffe-crf-v2.zip
train_DCL.prototxt
train_dcl.py
train_dclcrf.py
train_dss.py
train_dss_rosa.py
train_rfcn.py
train_rfcn_crf.py
utils.py

README.md

ROSA: Robust Salient Object Detection Against Adversarial Attacks

This is an implementation of 'ROSA: Robust Salient Object Detection Against Adversarial Attacks', published in IEEE Transactions on Cybernetics, 2019.

If you find this code helpful, please cite the following paper.

@article{li2019rosa,
  title={ROSA: Robust Salient Object Detection Against Adversarial Attacks},
  author={Li, Haofeng and Li, Guanbin and Yu, Yizhou},
  journal={IEEE Transactions on Cybernetics},
  pages={1--13},
  year={2019}
}

How to Use

Our method is a robust framework that protect a FCN from adversarial attacks. Our method consists of three components: segmentwise pixel shuffling, FCN backbone, CRF refinement. The code of pixel shuffling has not been released. The pairwise energy term is defined by the output of bilateral filter. The code of bilateral filtering has not been released. You may implement them by yourself.

The following is the model weights of a trained DSS+ROSA:

https://pan.baidu.com/s/1YV1NnBPbE_Usy57Y7C5fUQ password: lipc

You can’t perform that action at this time.