We have provided sample scripts in the 'sample_scripts' folder for two kinds of evaluations done in our work: 
(i) 'adv_attack.py': attacking students with adversarial images of the teacher (Section 4.2, Fig. 3).
(ii) 'crop_agreement': to test invariance of a model to random crops (Section C.4, appendix).

Since we are not releasing the pre-trained models yet, or the dataset, please consider these scripts as a pseudo-code rather than using it to reproduce the results. As mentioned towards the end of the Introduction section, we will release the complete code-base, including any datasets used, after publication.
