Generate Adversarial Attacked Image for Object Detection models using MMDetection
- You should put the files on /mmdetection/{here}
- This code use mmdetection 3.x version. DO NOT USE 2.x
- In this version, you can use only DAG method. I might update another methos later.
- This method very sensitive to changes in pixel values, so it stores the attacked image in [0,1] format rather than [0,255] format. For this purpose, the attacked images are stored in the form of a numpy array(.npy).
- This pipeline follows the following scheme