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:
Reimplement fully connected layer
Use unofficial evaluation and loss function
tf.data
Gabor convolution
ResNet
RCNN series
......