This repository incorporates different simple CNN implementations into different datasets that can be found from Keras datasets. We applied CNN architectures using both Keras API (functional) and Sequential model to make readers understanding how to create a model by using Tensorflow Keras in different ways. For your views and suggestions please contact me [email protected]

Notice that Fashion_mnist_Transfer_Learning.ipynb includes a ready model ('model_fashion-mnist_cnn_train2_epoch24.h5') can be downloaded from https://github.com/uchidama/Fashion-MNIST-Prediction-In-Keras/blob/master/model_fashion-mnist_cnn_train2_epoch24.h5.

We also publish Fashion_mnist_KerasAPI_model.ipynb that we build the same model architecture, applying Keras API, as used in the transfer learning and train our model to see its accuracy. Thus, one can see why transfer learning may be really important in case we want to fasten the process and increase the accuracy of our DL implementations. Since we got 76.% accuracy in Functional model (we built, compiled and trained with the training dataset) while model ready yielded 86.% accuracy.