Tensorflow 之 MNIST CNN實現並保存、加載模型

廢話不說，直接上代碼，註釋我覺得挺清晰了

# TensorFlow and tf.keras
import tensorflow as tf
from tensorflow import keras

# Helper libraries
import numpy as np
import matplotlib.pyplot as plt
import os

#download the data
mnist = keras.datasets.mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

class_names = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']

train_images = train_images / 255.0
test_images = test_images / 255.0

def create_model():
  # It's necessary to give the input_shape，or it will fail when you load the model
  # The error will be like : You are trying to load the 4 layer models to the 0 layer 
  model = keras.Sequential([
      keras.layers.Conv2D(32,[5,5], activation=tf.nn.relu,input_shape = (28,28,1)),
      keras.layers.MaxPool2D(),
      keras.layers.Conv2D(64,[7,7], activation=tf.nn.relu),
      keras.layers.MaxPool2D(),
      keras.layers.Flatten(),
      keras.layers.Dense(576, activation=tf.nn.relu),
      keras.layers.Dense(10, activation=tf.nn.softmax)
  ])

  model.compile(optimizer=tf.train.AdamOptimizer(), 
                loss='sparse_categorical_crossentropy',
                metrics=['accuracy'])
  
  return model

#reshape the shape before using it, for that the input of cnn is 4 dimensions
train_images = np.reshape(train_images,[-1,28,28,1])
test_images = np.reshape(test_images,[-1,28,28,1])


#train
model = create_model()                                                 
model.fit(train_images, train_labels, epochs=4)

#save the model
model.save('my_model.h5')

#Evaluate
test_loss, test_acc = model.evaluate(test_images, test_labels,verbose = 0)
print('Test accuracy:', test_acc)

模型保存後，自己手寫了幾張圖片，放在文件夾C:\pythonp\testdir2下，開始測試

#Load the model

new_model = keras.models.load_model('my_model.h5')
new_model.compile(optimizer=tf.train.AdamOptimizer(), 
                loss='sparse_categorical_crossentropy',
                metrics=['accuracy'])
new_model.summary()

#Evaluate

# test_loss, test_acc = new_model.evaluate(test_images, test_labels)
# print('Test accuracy:', test_acc)

#Predicte

mypath = 'C:\\pythonp\\testdir2'

def getimg(mypath):
    listdir = os.listdir(mypath)
    imgs = []
    for p in listdir:
        img = plt.imread(mypath+'\\'+p)
        # I save the picture that I draw myself under Windows, but the saved picture's
        # encode style is just opposite with the experiment data, so I transfer it with
        # this line. 
        img = np.abs(img/255-1)
        imgs.append(img[:,:,0])
    return np.array(imgs),len(imgs)

imgs = getimg(mypath)

test_images = np.reshape(imgs[0],[-1,28,28,1])

predictions = new_model.predict(test_images)

plt.figure()

for i in range(imgs[1]):
  c = np.argmax(predictions[i])
  plt.subplot(3,3,i+1)
  plt.xticks([])
  plt.yticks([])
  plt.imshow(test_images[i,:,:,0])
  plt.title(class_names[c])
plt.show()

測試結果

自己手寫的圖片截的時候要注意，空白部分儘量不要太大，否則測試結果就呵呵了