Convolution Layer Method

class NetworkBuilder:
    :
    :
    def attach_conv_layer(self, input_layer, output_size=32, feature_size=(5, 5), strides=[1, 1, 1, 1], padding='SAME',
                          summary=False):
        with tf.name_scope("Convolution") as scope:
            input_size = input_layer.get_shape().as_list()[-1]
            weights = tf.Variable(tf.random_normal([feature_size[0], feature_size[1], input_size, output_size]), name='conv_weights')
            if summary:
                tf.summary.histogram(weights.name, weights)
            biases = tf.Variable(tf.random_normal([output_size]),name='conv_biases')
            conv = tf.nn.conv2d(input_layer, weights, strides=strides, padding=padding)+biases
            return conv

Let’s see inside the method

The Pooling Layer

class NetworkBuilder:
    :
    :
    def attach_pooling_layer(self, input_layer, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME'):
        with tf.name_scope("Pooling") as scope:
            return tf.nn.max_pool(input_layer, ksize=ksize, strides=strides, padding=padding)

Activation Layer

class NetworkBuilder:
    :
    :
    def attach_relu_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.relu(input_layer)

    def attach_sigmoid_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.sigmoid(input_layer)

    def attach_softmax_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.softmax(input_layer)

Fully Connected Layer

Flatten Layer

class NetworkBuilder:
    :
    :
    def flatten(self, input_layer):
        with tf.name_scope("Flatten") as scope:
            input_size = input_layer.get_shape().as_list()
            new_size = input_size[-1]*input_size[-2]*input_size[-3]
            return tf.reshape(input_layer, [-1, new_size])

Finally The Fully Connected (Dense) Layer

class NetworkBuilder:
    :
    :
    def attach_dense_layer(self, input_layer, size, summary=False):
        with tf.name_scope("Dense") as scope:
            input_size = input_layer.get_shape().as_list()[-1]
            weights = tf.Variable(tf.random_normal([input_size, size]), name='dense_weigh')
            if summary:
                tf.summary.histogram(weights.name, weights)
            biases = tf.Variable(tf.random_normal([size]), name='dense_biases')
            dense = tf.matmul(input_layer, weights) + biases
            return dense

This Is The Final Network Builder Class

class NetworkBuilder:
    def __init__(self):
        pass

    def attach_conv_layer(self, input_layer, output_size=32, feature_size=(5, 5), strides=[1, 1, 1, 1], padding='SAME',
                          summary=False):
        with tf.name_scope("Convolution") as scope:
            input_size = input_layer.get_shape().as_list()[-1]
            weights = tf.Variable(tf.random_normal([feature_size[0], feature_size[1], input_size, output_size]), name='conv_weights')
            if summary:
                tf.summary.histogram(weights.name, weights)
            biases = tf.Variable(tf.random_normal([output_size]),name='conv_biases')
            conv = tf.nn.conv2d(input_layer, weights, strides=strides, padding=padding)+biases
            return conv

    def attach_relu_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.relu(input_layer)

    def attach_sigmoid_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.sigmoid(input_layer)

    def attach_softmax_layer(self, input_layer):
        with tf.name_scope("Activation") as scope:
            return tf.nn.softmax(input_layer)

    def attach_pooling_layer(self, input_layer, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],padding='SAME'):
        with tf.name_scope("Pooling") as scope:
            return tf.nn.max_pool(input_layer, ksize=ksize, strides=strides, padding=padding)

    def flatten(self, input_layer):
        with tf.name_scope("Flatten") as scope:
            input_size = input_layer.get_shape().as_list()
            new_size = input_size[-1]*input_size[-2]*input_size[-3]
            return tf.reshape(input_layer, [-1, new_size])

    def attach_dense_layer(self, input_layer, size, summary=False):
        with tf.name_scope("Dense") as scope:
            input_size = input_layer.get_shape().as_list()[-1]
            weights = tf.Variable(tf.random_normal([input_size, size]), name='dense_weigh')
            if summary:
                tf.summary.histogram(weights.name, weights)
            biases = tf.Variable(tf.random_normal([size]), name='dense_biases')
            dense = tf.matmul(input_layer, weights) + biases
            return dense

So What is Next?

• সুতরাং আসুন বর্গ নামের ঠিক নামে একটি ফাইলে এই কোড সংরক্ষণ করুন। যা NetworkBuilder.py হয়
• আমাদের পূর্ববর্তী টিউটোরিয়ালে ডেটাসেট জেনারেটর ক্লাসের দরকার হবে need ডেটা সেট  জেনারেটর শ্রেণীর মতো ঠিক একই ফাইলটির নামের সাথে অন্য কোডে সেই কোডটি সংরক্ষণ করুন, যা ছিল DataSetGenerator.py
• Put both of the files in the same.

• Unzip the Cat vs Dog train.zip in the same directory in train folder
• Create a new python file and let’s name it MyFirstCNNModel.py

Let’s Create a CNN Model in Tensorflow

Importing libraries

import tensorflow as tf 
from NetworkBuilder import NetworkBuilder 
from DataSetGenerator import DataSetGenerator,seperateData
import datetime 
import numpy as np 
import os

with tf.name_scope("Input") as scope:
    input_img = tf.placeholder(dtype='float', shape=[None, 128, 128, 3], name="input")

with tf.name_scope("Target") as scope:
    target_labels = tf.placeholder(dtype='float', shape=[None, 2], name="Targets")

Let’s Design the Model

nb = NetworkBuilder()

with tf.name_scope("ModelV2") as scope:
    model = input_img
    model = nb.attach_conv_layer(model, 32, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 32, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.attach_conv_layer(model, 64, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 64, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.attach_conv_layer(model, 128, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 128, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.flatten(model)
    model = nb.attach_dense_layer(model, 200, summary=True)
    model = nb.attach_sigmoid_layer(model)
    model = nb.attach_dense_layer(model, 32, summary=True)
    model = nb.attach_sigmoid_layer(model)
    model = nb.attach_dense_layer(model, 2)
    prediction = nb.attach_softmax_layer(model)

Now we will create the optimization and accuracy blocks

with tf.name_scope("Optimization") as scope:
    global_step = tf.Variable(0, name='global_step', trainable=False)
    cost = tf.nn.softmax_cross_entropy_with_logits(logits=model, labels=target_labels)
    cost = tf.reduce_mean(cost)
    tf.summary.scalar("cost", cost)

    optimizer = tf.train.AdamOptimizer().minimize(cost,global_step=global_step)

with tf.name_scope('accuracy') as scope:
    correct_pred = tf.equal(tf.argmax(prediction, 1), tf.argmax(target_labels, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

Visualize in Tensorboard

We completed The CNN Image Classifier Model, let’s train it

# we need to run this only once
#seperateData("./train")
dg = DataSetGenerator("./train")


epochs = 10
batchSize = 10

saver = tf.train.Saver()
model_save_path="./saved model v2/"
model_name='model'

with tf.Session() as sess:
    summaryMerged = tf.summary.merge_all()

    filename = "./summary_log/run" + datetime.datetime.now().strftime("%Y-%m-%d--%H-%M-%s")
    # setting global steps
    tf.global_variables_initializer().run()

    if os.path.exists(model_save_path+'checkpoint'):
        # saver = tf.train.import_meta_graph('./saved '+modelName+'/model.ckpt.meta')
        saver.restore(sess, tf.train.latest_checkpoint(model_save_path))
    writer = tf.summary.FileWriter(filename, sess.graph)

    for epoch in range(epochs):
        batches = dg.get_mini_batches(batchSize,(128,128), allchannel=False)
        for imgs ,labels in batches:
            imgs=np.divide(imgs, 255)
            error, sumOut, acu, steps,_ = sess.run([cost, summaryMerged, accuracy,global_step,optimizer],
                                            feed_dict={input_img: imgs, target_labels: labels})
            writer.add_summary(sumOut, steps)
            print("epoch=", epoch, "Total Samples Trained=", steps*batchSize, "err=", error, "accuracy=", acu)
            if steps % 100 == 0:
                print("Saving the model")
                saver.save(sess, model_save_path+model_name, global_step=steps)

import tensorflow as tf
from NetworkBuilder import NetworkBuilder
from DataSetGenerator import DataSetGenerator, seperateData
import datetime
import numpy as np
import os


with tf.name_scope("Input") as scope:
    input_img = tf.placeholder(dtype='float', shape=[None, 128, 128, 1], name="input")

with tf.name_scope("Target") as scope:
    target_labels = tf.placeholder(dtype='float', shape=[None, 2], name="Targets")

with tf.name_scope("Keep_prob_input") as scope:
    keep_prob = tf.placeholder(dtype='float',name='keep_prob')

nb = NetworkBuilder()

with tf.name_scope("ModelV2") as scope:
    model = input_img
    model = nb.attach_conv_layer(model, 32, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 32, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.attach_conv_layer(model, 64, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 64, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.attach_conv_layer(model, 128, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_conv_layer(model, 128, summary=True)
    model = nb.attach_relu_layer(model)
    model = nb.attach_pooling_layer(model)

    model = nb.flatten(model)
    model = nb.attach_dense_layer(model, 200, summary=True)
    model = nb.attach_sigmoid_layer(model)
    model = nb.attach_dense_layer(model, 32, summary=True)
    model = nb.attach_sigmoid_layer(model)
    model = nb.attach_dense_layer(model, 2)
    prediction = nb.attach_softmax_layer(model)


with tf.name_scope("Optimization") as scope:
    global_step = tf.Variable(0, name='global_step', trainable=False)
    cost = tf.nn.softmax_cross_entropy_with_logits(logits=model, labels=target_labels)
    cost = tf.reduce_mean(cost)
    tf.summary.scalar("cost", cost)
    optimizer = tf.train.AdamOptimizer().minimize(cost, global_step=global_step)

with tf.name_scope('accuracy') as scope:
    correct_pred = tf.equal(tf.argmax(prediction, 1), tf.argmax(target_labels, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

dg = DataSetGenerator("./train")


epochs = 10
batchSize = 10

saver = tf.train.Saver()
model_save_path="./saved model v2/"
model_name='model'

with tf.Session() as sess:
    summaryMerged = tf.summary.merge_all()

    filename = "./summary_log/run" + datetime.datetime.now().strftime("%Y-%m-%d--%H-%M-%s")
    # setting global steps
    tf.global_variables_initializer().run()

    if os.path.exists(model_save_path+'checkpoint'):
        # saver = tf.train.import_meta_graph('./saved '+modelName+'/model.ckpt.meta')
        saver.restore(sess, tf.train.latest_checkpoint(model_save_path))
    writer = tf.summary.FileWriter(filename, sess.graph)

    for epoch in range(epochs):
        batches = dg.get_mini_batches(batchSize,(128,128), allchannel=False)
        for imgs ,labels in batches:
            imgs=np.divide(imgs, 255)
            error, sumOut, acu, steps,_ = sess.run([cost, summaryMerged, accuracy,global_step,optimizer],
                                            feed_dict={input_img: imgs, target_labels: labels})
            writer.add_summary(sumOut, steps)
            print("epoch=", epoch, "Total Samples Trained=", steps*batchSize, "err=", error, "accuracy=", acu)
            if steps % 100 == 0:
                print("Saving the model")
                saver.save(sess, model_save_path+model_name, global_step=steps)

class NetworkBuilder:
    :
    :
    def attach_inception_module(self,input_layer, output_size):
        output_size_road1 = int(output_size*0.2)
        road1 = self.attach_conv_layer(input_layer=input_layer, output_size=output_size_road1,
                                       feature_size=(1, 1))
        road1 = self.attach_relu_layer(road1)

        output_size_road2 = int(output_size * 0.3)
        road2 = self.attach_conv_layer(input_layer=input_layer, output_size=output_size_road2,
                                       feature_size=(1, 1))
        road2 = self.attach_relu_layer(road2)
        road2 = self.attach_conv_layer(input_layer=road2, output_size=output_size_road2,
                                       feature_size=(3, 3))

        output_size_road3 = int(output_size * 0.3)
        road3 = self.attach_conv_layer(input_layer=input_layer, output_size=output_size_road3,
                                       feature_size=(1, 1))
        road3 = self.attach_relu_layer(road3)
        road3 = self.attach_conv_layer(input_layer=road3, output_size=output_size_road2,
                                       feature_size=(5, 5))

        output_size_road4 = output_size-(output_size_road1 + output_size_road2 + output_size_road3)
        road4 = self.attach_pooling_layer(input_layer=input_layer, strides=[1, 1, 1, 1])
        road4 = self.attach_conv_layer(input_layer=road4, output_size=output_size_road4,
                                       feature_size=(1, 1))

        with tf.name_scope("FilterConcat") as scope:
            concat = tf.concat([road1, road2, road3, road4], axis=3, name="FilterConcat")
            concat = self.attach_relu_layer(concat)
        return concat