An attempt to reuse layers and pretrained weights of models from keras applications

tl_experiment.py

"""
An attempt to reuse layers and pretrained weights of models from keras applications
The background to this attempt is here: https://www.tensorflow.org/tutorials/images/transfer_learning
"""
from __future__ import absolute_import, division, print_function

import os

import tensorflow as tf
from tensorflow import keras

print("\n\nTensorFlow version is ", tf.__version__)

image_size = 160

IMG_SHAPE = (image_size, image_size, 3)

# Create the base model from the pre-trained model MobileNet V2
# input_tensor = keras.Input(shape=IMG_SHAPE)
base_model = tf.keras.applications.VGG19(input_shape=IMG_SHAPE,
                                          # input_tensor=input_tensor,
                                          include_top=False, weights='imagenet')

base_model.trainable = True

# Sequential model approach:
seq_model = keras.Sequential([
  # keras.Input(shape=IMG_SHAPE), # we dont need this line for sequential model build
  base_model.layers[0],
  base_model.layers[1],
  base_model.layers[2],
  keras.layers.Conv2D(64,(3,3), activation='relu')
])
# print out model summary
seq_model.summary()
# print(seq_model.layers[0].weights[0][0])
print(seq_model.layers[0].input)
print("\n\n")

# Functional api approach
inputs = keras.Input(shape=IMG_SHAPE)
# inputs = base_model.layers[0]
m_layer = base_model.layers[1](inputs)
m_layer = base_model.layers[2](m_layer)
outputs = keras.layers.Conv2D(64,(3,3), activation='relu')(m_layer)


api_model = keras.Model(inputs=inputs, outputs=outputs)

# print out model summary
api_model.summary()

# print(api_model.layers[0].weights)

# Both approaches turn out the same

"""
You can find details of each model in the keras applications here:
https://github.com/keras-team/keras-applications
"""