CIFAR-100--Knowledge-Distillation-with-Augmented-Data

Knowledge distillation using CIFAR 100

Paper : https://arxiv.org/abs/1503.02531

Teacher Model

Resnet 20 (you can change with 50 or others)

Student Model

MobileNet or CNN - LSTM

CNN - LSTM model

Data

ImageNet Dataset : http://www.image-net.org/

How to get dataset?

Link = http://hpkim0512.blogspot.com/2017/12/sdf.html?m=1

Training Method

Stop Training if overfitting(validation set accuracy doesn't increase) happens more than 50 epochs

Programming Language

Python 3.6 Tensorflow, keras

OS dependency

windows 10, ubuntu linux

Result

Resnet - 20's Training result

Training result with spatial complexity

Training result with time complexity