Chapter 5: Convolutional Neural Networks for Computer Vision

Activity 5: Making Use of Data Augmentation to Classify correctly Images of Flowers

Solution

1. Open your Google Colab interface.

   Note:

   You will need to mount your drive using the Dataset folder, and accordingly set the path to continue ahead.

   import numpyasnp

   classes=['daisy','dandelion','rose','sunflower','tulip']

   X=np.load("Dataset/flowers/%s_x.npy"%(classes[0]))

   y=np.load("Dataset/flowers/%s_y.npy"%(classes[0]))

   print(X.shape)

   forflowerinclasses[1:]:

   X_aux=np.load("Dataset/flowers/%s_x.npy"%(flower))

   y_aux=np.load("Dataset/flowers/%s_y.npy"%(flower))

   print(X_aux.shape)

   X=np.concatenate((X,X_aux),axis=0)

   y=np.concatenate((y,y_aux),axis=0)

   print(X.shape)

   print(y.shape)

2. To output some samples from the dataset:

   import random

   random.seed(42)

   from matplotlib import pyplot as plt

   import cv2

   for idx in range(5):

   rnd_index = random.randint(0, 4000)

   plt.subplot(1,5,idx+1),plt.imshow(cv2.cvtColor(X[rnd_index],cv2.COLOR_BGR2RGB))

   plt.xticks([]),plt.yticks([])

   plt.savefig("flowers_samples.jpg", bbox_inches='tight')

   plt.show()

   The output is as follows:

   

   Figure 5.23: Samples from the dataset

3. Now, we will normalize and perform one-hot encoding:

   from keras import utils as np_utils

   X = (X.astype(np.float32))/255.0

   y = np_utils.to_categorical(y, len(classes))

   print(X.shape)

   print(y.shape)

4. Splitting the training and testing set:

   from sklearn.model_selection import train_test_split

   x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

   input_shape = x_train.shape[1:]

   print(x_train.shape)

   print(y_train.shape)

   print(x_test.shape)

   print(y_test.shape)

   print(input_shape)

5. Import libraries and build the CNN:

   from keras.models import Sequential

   from keras.callbacks import ModelCheckpoint

   from keras.layers import Input, Dense, Dropout, Flatten

   from keras.layers import Conv2D, Activation, BatchNormalization

   def CNN(input_shape):

   model = Sequential()

   model.add(Conv2D(32, kernel_size=(5, 5), padding='same', strides=(2,2), input_shape=input_shape))

   model.add(Activation('relu'))

   model.add(BatchNormalization())

   model.add(Dropout(0.2))

   model.add(Conv2D(64, kernel_size=(3, 3), padding='same', strides=(2,2)))

   model.add(Activation('relu'))

   model.add(BatchNormalization())

   model.add(Dropout(0.2))

   model.add(Conv2D(128, kernel_size=(3, 3), padding='same', strides=(2,2)))

   model.add(Activation('relu'))

   model.add(BatchNormalization())

   model.add(Dropout(0.2))

   model.add(Conv2D(256, kernel_size=(3, 3), padding='same', strides=(2,2)))

   model.add(Activation('relu'))

   model.add(BatchNormalization())

   model.add(Dropout(0.2))

   model.add(Flatten())

   model.add(Dense(512))

   model.add(Activation('relu'))

   model.add(BatchNormalization())

   model.add(Dropout(0.5))

   model.add(Dense(5, activation = "softmax"))

   return model

6. Declare ImageDataGenerator:

   from keras.preprocessing.image import ImageDataGenerator

   datagen = ImageDataGenerator(

   rotation_range=10,

   zoom_range = 0.2,

   width_shift_range=0.2,

   height_shift_range=0.2,

   shear_range=0.1,

   horizontal_flip=True

   )

7. We will now train the model:

   datagen.fit(x_train)

   model = CNN(input_shape)

   model.compile(loss='categorical_crossentropy', optimizer='Adadelta', metrics=['accuracy'])

   ckpt = ModelCheckpoint('Models/model_flowers.h5', save_best_only=True,monitor='val_loss', mode='min', save_weights_only=False)

   //{…}##the detailed code can be found on Github##

   model.fit_generator(datagen.flow(x_train, y_train,

   batch_size=32),

   epochs=200,

   validation_data=(x_test, y_test),

   callbacks=[ckpt],

   steps_per_epoch=len(x_train) // 32,

   workers=4)

8. After which, we will evaluate the model:

   from sklearn import metrics

   model.load_weights('Models/model_flowers.h5')

   y_pred = model.predict(x_test, batch_size=32, verbose=0)

   y_pred = np.argmax(y_pred, axis=1)

   y_test_aux = y_test.copy()

   y_test_pred = list()

   for i in y_test_aux:

   y_test_pred.append(np.argmax(i))

   //{…}

   ##the detailed code can be found on Github##

   print (y_pred)

   # Evaluate the prediction

   accuracy = metrics.accuracy_score(y_test_pred, y_pred)

   print('Acc: %.4f' % accuracy)

9. The accuracy achieved is 91.68%.
10. Try the model with unseen data:

    classes = ['daisy','dandelion','rose','sunflower','tulip']

    images = ['sunflower.jpg','daisy.jpg','rose.jpg','dandelion.jpg','tulip .jpg']

    model.load_weights('Models/model_flowers.h5')

    for number in range(len(images)):

    imgLoaded = cv2.imread('Dataset/testing/%s'%(images[number]))

    img = cv2.resize(imgLoaded, (150, 150))

    img = (img.astype(np.float32))/255.0

    img = img.reshape(1, 150, 150, 3)

    plt.subplot(1,5,number+1),plt.imshow(cv2.cvtColor(imgLoaded,cv2.COLOR_BGR2RGB))

    plt.title(np.argmax(model.predict(img)[0]))

    plt.xticks([]),plt.yticks([])

    plt.show()

    Output will look like this:

Figure 5.24: Prediction of roses result from Activity05

Note:

The detailed code for this activity can be found on GitHub - https://github.com/PacktPublishing/Artificial-Vision-and-Language-Processing-for-Robotics/blob/master/Lesson05/Activity05/Activity05.ipynb