- download the pre-trained classifier on GTSRD from https://drive.google.com/drive/folders/1l4s6XLx9PVRSUl1ZXUguK8Wa_1pGXkBH?usp=sharing
- collect digital and physical images dataset of GTSRD or ImageNet , put them in
imagefolder like this:
image
|
|--label1
| |--TrainA(digital)
| |--TrainB(physical)
|
|--label2
| |--TrainA(digital)
| |--TrainB(physical)
|--label3
| |--TrainA(digital)
| |--TrainB(physical)
.
.
.
-
run
./run.shto train the GAN networks -
run
./run_test.shto finetune the GAN on unseen images and unseen classifiers -
evaluate the generated adversarial images in checkpoints folder