Overview

This is just a example to train Fashion-MNIST by Convolutional Neural Network for learning TensorFlow 2.0.

Description

• This is an assignment of Deep Learning basic class arranged a little.
• Training Fashion-MNIST by CNN on Google Colaboratory with TensorFlow 2.0 Alpha.
• Data is augmented by ImageDataGenerator of Keras and the effectiveness of data augmentation is shown.

Attention

Fashion-MNIST-by-CNN-with-data-augmentation.ipynb file can not be opened correctly, I'm not sure about the reason though. However, you can see the code in Google Colaboratory. So please take a look at How-to-open-ipynb-on-Google-Colaboratory and open it in Google Colaboratory.

Dataset

This dataset can be found here.

Model

4 Layer CNN with Max Pooling, Batch Normalization and Dropout.

Accuracy

• with Data Augmentation
  92.2% after training in 80 epochs
• witout Data Augmentation
  91.8% after training in 80 epochs
  

How to open it on Google Colaboratory

Please take a look at How-to-open-ipynb-on-Google-Colaboratory.