Tensorflow models

In this repository you can find Tensorflow-keras implementations of ResNeXt and dual path network model architectures.

Instructions

In order to be able to use these models tensorflow version at least 2.1 needs to be installed on your environment. Follow the instructions at https://www.tensorflow.org/install to see available options for installation.

In order to load the models onto your code follow the steps listed below.

Open a terminal and paste the following code to make a local copy of this repository:

# Clone the repository onto a local folder
git clone https://github.com/PathofData/Tensorflow-models.git

Then inside your python code import the model of your choice:

# Import the DPN module
from DPN50 import DPN50

# Initiallize a new model instance where the image dimension
# is (224, 224, 3) and we wish to predict 1000 classes
vision_model = DPN50(include_top=True,
                     weights=None,
                     input_tensor=None,
                     input_shape=(224, 224, 3),
                     pooling=None,
                     classes=1000)

Optionally print a summary of the model:

# Print each layere input, output shapes and number of parameters
vision_model.summary()

Once you have prepared a dataset for training use this model like how one would use any instance of keras models

# Compile the model with an optimizer and a loss function
vision_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy')

# Train the model on some data
vision_model.fit(X_train, y_train)