CReLU layer with TensorFlow2.
A simple model applying CReLU activation to an input layer, with save/load function
from tensorflow.keras.layers import Input
from tensorflow.keras.models import Model, load_model
from crelu import CReLU
ip = Input(shape=(3, 3, 6))
x = CReLU()(ip)
model = Model(ip, x)
model.summary()
model.save( 'model.h5' )
print( '*'*80 )
nm = load_model( 'model.h5' )
print( 'new model loaded successfully' )produces
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 3, 3, 6)] 0
_________________________________________________________________
c_re_lu (CReLU) (None, 3, 3, 12) 0
=================================================================
Total params: 0
Trainable params: 0
Non-trainable params: 0
_________________________________________________________________
********************************************************************************
new model loaded successfully
Before loading a model with CReLU layer(s), make sure from crelu import CReLU has been executed.
Shang, Wenling, Kihyuk Sohn, Diogo Almeida, and Honglak Lee. “Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear Units.” ArXiv:1603.05201 [Cs], July 19, 2016. http://arxiv.org/abs/1603.05201.
AGPL-3