[PYTHON] Judge whether it is my child from the picture of Shiba Inu by deep learning (1)

Step 3 Copy the specified number of Shiba Inu photos (60 each) to the train, test, and validation folders.

(1) Set the file path for copying the image file and create a new folder to use.

#Loading the required libraries
import os, shutil

#Define the required file path
# cd '/content/drive/'My Drive/'Colab Notebooks'Change the current directory to the working folder in
%cd '/content/drive/'My Drive/Colab Notebooks/Self_Study/02_mydog_or_otherdogs

# original_Setting the file path of the data folder
original_dataset_dir = 'original_data'
# original_data(Definition name'original_dataset_dir')Set the following two folder paths in.
original_mydog_dir = 'original_data/mydog'
original_otherdogs_dir = 'original_data/otherdogs'

# use_Data file path settings
base_dir = 'use_data'

# use_data folder(Definition name'base_dir')Set the following three folder paths in. Create a folder.
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)
validation_dir = os.path.join(base_dir, 'validation')
os.mkdir(validation_dir)
test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)

#In the train folder'mydog'With folders'otherdogs'Set the folder path. Create a folder.
train_mydog_dir = os.path.join(train_dir, 'mydog')
os.mkdir(train_mydog_dir)
train_otherdogs_dir = os.path.join(train_dir, 'otherdogs')
os.mkdir(train_otherdogs_dir)

#In the validation folder'mydog'With folders'otherdogs'Set the folder path. Create a folder.
validation_mydog_dir = os.path.join(validation_dir, 'mydog')
os.mkdir(validation_mydog_dir)
validation_otherdogs_dir = os.path.join(validation_dir, 'otherdogs')
os.mkdir(validation_otherdogs_dir)

#In the test folder'mydog'With folders'otherdogs'Set the folder path. Create a folder.
test_mydog_dir = os.path.join(test_dir, 'mydog')
os.mkdir(test_mydog_dir)
test_otherdogs_dir = os.path.join(test_dir, 'otherdogs')
os.mkdir(test_otherdogs_dir)

(2) Make a file copy and check the number of copied files.

#train for train_mydog_Copy 30 mydog files to dir
fnames = ['mydog{}.jpg'.format(i) for i in range(30)]
for fname in fnames:
    src = os.path.join(original_mydog_dir, fname)
    dst = os.path.join(train_mydog_dir, fname)
    shutil.copyfile(src, dst)

#validation for validation_mydog_Copy 20 mydog files to dir
fnames = ['mydog{}.jpg'.format(i) for i in range(30,50)]
for fname in fnames:
    src = os.path.join(original_mydog_dir, fname)
    dst = os.path.join(validation_mydog_dir, fname)
    shutil.copyfile(src, dst)

#test for test_mydog_Copy 10 mydog files to dir
fnames = ['mydog{}.jpg'.format(i) for i in range(50,60)]
for fname in fnames:
    src = os.path.join(original_mydog_dir, fname)
    dst = os.path.join(test_mydog_dir, fname)
    shutil.copyfile(src, dst)

#train for train_otherdogs_Copy 30 otherdogs files to dir
fnames = ['otherdogs{}.jpg'.format(i) for i in range(30)]
for fname in fnames:
    src = os.path.join(original_otherdogs_dir, fname)
    dst = os.path.join(train_otherdogs_dir, fname)
    shutil.copyfile(src, dst)

#validation for validation_otherdogs_Copy 20 otherdogs files to dir
fnames = ['otherdogs{}.jpg'.format(i) for i in range(30, 50)]
for fname in fnames:
    src = os.path.join(original_otherdogs_dir, fname)
    dst = os.path.join(validation_otherdogs_dir, fname)
    shutil.copyfile(src, dst)

#test for test_otherdogs_Copy 10 otherdogs files to dir
fnames = ['otherdogs{}.jpg'.format(i) for i in range(50, 60)]
for fname in fnames:
    src = os.path.join(original_otherdogs_dir, fname)
    dst = os.path.join(test_otherdogs_dir, fname)
    shutil.copyfile(src, dst)

#Check the number of files stored in each folder
print('total training mydog images:', len(os.listdir(train_mydog_dir)))
print('total training otherdogs images:', len(os.listdir(train_otherdogs_dir)))
print('total validation mydog images:', len(os.listdir(validation_mydog_dir)))
print('total validation otherdogs images:', len(os.listdir(validation_otherdogs_dir)))
print('total test mydog images:', len(os.listdir(test_mydog_dir)))
print('total test otherdogs images:', len(os.listdir(test_otherdogs_dir)))

The number of files is displayed as follows.

Step 4 Model construction

(1) Import keras and build a model as follows.

The size of the input image is 320 * 320 in the original file, but this time the size of the input is read as 150 * 150.

from keras import layers
from keras import models

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.summary()

The model configuration is displayed like this.

(2) Set the loss function (binary crossentropy), optimizer (RMS Prop), and monitoring index (accuracy), and compile the model.

from keras import optimizers

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])

(3) Convert the image file to a tensor using the keras library ImageDataGenerator.

No data augmentation was implemented this time.

from keras.preprocessing.image import ImageDataGenerator

# rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # target directory
        train_dir,
        # resized to 150x150
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

Step 5 Learning

(1) Train the model using the batch generator.

history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=50)

When learning starts, the display will look like this, and the calculation will proceed. (It will take a while)

(2) Save the model after training.

model.save('mydog_or_otherdogs_01a.h5')

Step 6 result

(1) Display the learning result as a graph

import matplotlib.pyplot as plt

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()

plt.show()

Result graph

First of all, accuracy

--The accuracy of traning has reached 100% immediately from the beginning. --The accuracy of validation is around 70% to 85%, and the accuracy does not improve any more. You can see the features such as. Unfortunately, we are only looking at the characteristics that are valid only for train data, and we cannot say that we have a clear characteristic of our dog that is valid for other than train data.

On the other hand, about loss

--Training loss is close to 0 from the beginning. ――Validation loss does not decrease as the number of times increases, but increases upward.

From the above, we are overfitting the train data. After all, the number of data was originally small, so it seems that the result is unavoidable.

(2) Application to test data

Let's apply this model to the test data with the following code to see the classification accuracy. ** Note) ImageDataGenerator will fail without conversion if there are no two or more subfolders in the target folder. ** (In this case you need two subfolders, mydog and otherdogs) </ font>

test_datagen = ImageDataGenerator(rescale=1./255)
test_generator = test_datagen.flow_from_directory(
    test_dir,
    target_size=(150,150),
    batch_size=50,
    class_mode='binary')

test_loss, test_acc = model.evaluate_generator(test_generator, steps=50)
print('test loss:', test_loss)
print('test acc:', test_acc)

#Inflated train data
train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,)

#Validation data is not padded
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # Target directory
        train_dir,
        # resized to 150x150
        target_size=(150, 150),
        batch_size=32,
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=32,
        class_mode='binary')

history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=100,
      validation_data=validation_generator,
      validation_steps=50)

model.save('mydog_or_otherdogs_01b.h5')