This is an example for tensorflow 2.0-alpha

In version 2.0, many features of tensorflow are deleted or changed, such as tf.Session. And in new version, Keras became the core function, which makes the code style quite different from before.

In this project, I will implement various layers and models. And I will try not to rely on advanced APIs to make it easy to change the details of the model.

Models and layers are as follows:

DONE:

Reimplement fully connected layer

Reimplement convolution layer

Use unofficial evaluation and loss function

Use unofficial train function

Batch normalization

VGG-16

LSTM

TODO:

tf.data

Gabor convolution

ResNet

RCNN series

......