python - How to get better predictions by data augmentation on CNN model for Image Classification

Question

I’m a beginner. I try Image Classification to my photos by using CNN model for on MNIST. To get better predictions, I did data augmentation to this, but I couldn’t improve it. Something wrong? And I don’t understand why CASE A on the following table had the wrong prediction to the number “4” comparing with CASE B, even though it has almost same accuracy at each time. Please give me some advice.

my 10 photos (.png, 28x28pixel)

1. Change the number of MNIST data to 5 cases
2. Data augmentation No or Yes to 5 cases
3. Predict my 10 photos by using learned model

# import library
from tensorflow import keras
from tensorflow.keras import datasets, layers, models
from tensorflow.keras.layers import Dense, Conv2D, Flatten, Dropout, MaxPooling2D
from tensorflow.keras.models import Sequential
from tensorflow.keras.preprocessing.image import ImageDataGenerator

# MNIST
mnist=keras.datasets.mnist
(x_train,y_train),(x_test,y_test)=mnist.load_data()
#(x_train,y_train),(x_test,y_test)=(x_train[:80],y_train[:80]),(x_test[:20], y_test[:20])
#(x_train,y_train),(x_test,y_test)=(x_train[:160],y_train[:160]),(x_test[:40], y_test[:40])
#(x_train,y_train),(x_test,y_test)=(x_train[:800],y_train[:800]),(x_test[:200], y_test[:200])
#(x_train,y_train),(x_test,y_test)=(x_train[:8000],y_train[:8000]),(x_test[:2000], y_test[:2000])
x_train=x_train.reshape(x_train.shape[0],28,28,1)
x_test=x_test.reshape(x_test.shape[0],28,28,1)
x_train=x_train/255
x_test=x_test/255
print("x_train",x_train.shape)
print("x_test",x_test.shape)

# Convolutional Neural Networks
model = Sequential()
model.add(Conv2D(16,(3,3),padding='same',input_shape=(28,28,1),activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Conv2D(128,(3,3),activation='relu'))
model.add(Conv2D(256,(3,3),activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(128,activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(10,activation='softmax'))
model.summary()

# model compile and learn
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(x_train,y_train,epochs=5)

# evoluate for test data
loss,acc=model.evaluate(x_test,y_test,verbose=2)
print('accuracy:',acc)

# data augmentation
datagen=ImageDataGenerator(rescale=1/255,
                           width_shift_range=0.01,
                           height_shift_range=0.025,                                                    
                           zoom_range=0.05)

# learn
history=model.fit_generator(
datagen.flow(x_train,y_train,batch_size=64),
steps_per_epoch=60,
epochs=40,
validation_data=(x_test,y_test),
validation_steps=5,
verbose=1)

# evoluate for test data
loss,acc=model.evaluate(x_test,y_test,verbose=2)
print('accuracy:',acc)

question from:https://stackoverflow.com/questions/65889069/how-to-get-better-predictions-by-data-augmentation-on-cnn-model-for-image-classi

Answer 1

Waitting for answers