We use code from the and modified it for our experiments.

To install requirements:

pip install -r requirements.txt

Networks can be trained on 16-class ImageNet dataset using the following command:

python train.py --tag trainSCAN_MODE_COLOR --mode MODE --data imagenet --class_num 16 --epoch 200

where noise is 'color', 'gray', 'noise' or 'blur'.

Trained networks can be tested on the same images used to test humans, using the following command:

python test.py --tag testSCAN_MODE_COLOR --mode MODE --data imagenet --class_num 16 --load_path LOAD_PATH

where MODE is 'color', 'gray', 'noise' or 'blur' and LOAD_PATH is the path to the saved model.