Scribd
Upload
5 views3 pages

CNN Model Training with Keras

The document outlines a Python script that utilizes TensorFlow to create and train a convolutional neural network (CNN) for image processing tasks. It includes functions for model creation, data loading, preprocessing, and training, along with parameters for optimization. The trained model is saved for future use after training and validation.

Uploaded by

AlexandruVlad
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
5 views3 pages

CNN Model Training with Keras

The document outlines a Python script that utilizes TensorFlow to create and train a convolutional neural network (CNN) for image processing tasks. It includes functions for model creation, data loading, preprocessing, and training, along with parameters for optimization. The trained model is saved for future use after training and validation.

Uploaded by

AlexandruVlad
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd

import tensorflow as tf

import numpy as np
import pickle
from [Link] import shuffle
from sklearn.model_selection import train_test_split
from [Link] import Sequential
from [Link] import Dropout, UpSampling2D
from [Link] import Conv2DTranspose, Conv2D, MaxPooling2D
from [Link] import ImageDataGenerator
from [Link] import BatchNormalization
import [Link] as plt

[Link].set_verbosity([Link])

def create_model(input_shape, pool_size):


# Create the actual neural network here
model = Sequential()
# Normalizes incoming inputs. First layer needs the input shape to work
[Link](BatchNormalization(input_shape=input_shape))

# Below layers were re-named for easier reading of model summary; this not
necessary
# Conv Layer 1
[Link](Conv2D(8, (3, 3), strides=(1, 1), activation='relu', name='Conv1'))

# Conv Layer 2
[Link](Conv2D(16, (3, 3), strides=(1, 1), activation='relu', name='Conv2'))

# Pooling 1
[Link](MaxPooling2D(pool_size=pool_size))

# Conv Layer 3
[Link](Conv2D(16, (3, 3), strides=(1, 1), activation='relu', name='Conv3'))
[Link](Dropout(0.2))

# Conv Layer 4
[Link](Conv2D(32, (3, 3), strides=(1, 1), activation='relu', name='Conv4'))
[Link](Dropout(0.2))

# Conv Layer 5
[Link](Conv2D(32, (3, 3), strides=(1, 1), activation='relu', name='Conv5'))
[Link](Dropout(0.2))

# Pooling 2
[Link](MaxPooling2D(pool_size=pool_size))

# Conv Layer 6
[Link](Conv2D(64, (3, 3), strides=(1, 1), activation='relu', name='Conv6'))
[Link](Dropout(0.2))

# Conv Layer 7
[Link](Conv2D(64, (3, 3), strides=(1, 1), activation='relu', name='Conv7'))
[Link](Dropout(0.2))

# Pooling 3
[Link](MaxPooling2D(pool_size=pool_size))

# Upsample 1
[Link](UpSampling2D(size=pool_size))

# Deconv 1
[Link](Conv2DTranspose(64, (3, 3), strides=(1, 1), activation='relu',
name='Deconv1'))
[Link](Dropout(0.2))

# Deconv 2
[Link](Conv2DTranspose(64, (3, 3), strides=(1, 1), activation='relu',
name='Deconv2'))
[Link](Dropout(0.2))

# Upsample 2
[Link](UpSampling2D(size=pool_size))

# Deconv 3
[Link](Conv2DTranspose(32, (3, 3), strides=(1, 1), activation='relu',
name='Deconv3'))
[Link](Dropout(0.2))

# Deconv 4
[Link](Conv2DTranspose(32, (3, 3), strides=(1, 1), activation='relu',
name='Deconv4'))
[Link](Dropout(0.2))

# Deconv 5
[Link](Conv2DTranspose(16, (3, 3), strides=(1, 1), activation='relu',
name='Deconv5'))
[Link](Dropout(0.2))

# Upsample 3
[Link](UpSampling2D(size=pool_size))

# Deconv 6
[Link](Conv2DTranspose(16, (3, 3), strides=(1, 1), activation='relu',
name='Deconv6'))

# Final layer - only including one channel so 1 filter


[Link](Conv2DTranspose(1, (3, 3), strides=(1, 1), activation='relu',
name='Final'))

return model

def main():
# Load training images
train_images = [Link](open("full_CNN_train.p", "rb"))

# Load image labels


labels = [Link](open("full_CNN_labels.p", "rb"))

# Make into arrays as the neural network wants these


train_images = [Link](train_images)
labels = [Link](labels)

# Normalize labels - training images get normalized to start in the network


labels = labels / 255

# Shuffle images along with their labels, then split into training/validation
sets
train_images, labels = shuffle(train_images, labels)
# Test size may be 10% or 20%
X_train, X_val, y_train, y_val = train_test_split(train_images, labels,
test_size=0.1)

# Batch size, epochs and pool size below are all paramaters to fiddle with for
optimization
batch_size = 128
epochs = 10
pool_size = (2, 2)
input_shape = X_train.shape[1:]

# Create the neural network


model = create_model(input_shape, pool_size) # type: Sequential

# Using a generator to help the model use less data


# Channel shifts help with shadows slightly
datagen = ImageDataGenerator(channel_shift_range=0.2)
[Link](X_train)

# Compiling and training the model


[Link](optimizer='Adam', loss='mean_squared_error',
metrics=['accuracy'])
model.fit_generator([Link](X_train, y_train, batch_size=batch_size),
steps_per_epoch=len(X_train) / batch_size, epochs=epochs,
validation_data=(X_val, y_val))

# Freeze layers since training is done


[Link] = False
[Link](optimizer='Adam', loss='mean_squared_error')

# Save model architecture and weights


[Link]('full_CNN_model_test.h5')

# Show summary of model


[Link]()

if __name__ == '__main__':
main()

You might also like