Teacher AE as here: https://openaccess.thecvf.com/content/CVPR2022W/WiCV/papers/Schneider_Autoencoders_-_A_Comparative_Analysis_in_the_Realm_of_Anomaly_CVPRW_2022_paper.pdf
Teacher free parameters: 155,981
Teacher inference time/img on CPU: 32 ms
Student free parameters: 216
Student inference time/img on CPU: 1.5 ms
Compression factor in free parameters: 722
Compression factor in FLOPS: 300
In folders, FASHION_MNIST and MNIST, you will find the teachers trained for each label and corresponding students, distinguished by their size measured in number of free parameters.
In folders with 'student_normals', you will find incomplete analysis only using a fraction of normal digits for training student.
In folder drawio_files, you find files to upload into https://app.diagrams.net/ to change architecture plots.