VGG
作者:互联网
vgg代码如下:
1 import tensorflow as tf 2 from tensorflow import keras 3 from tensorflow.keras import layers, regularizers 4 import numpy as np 5 import os 6 import cv2 7 import matplotlib.pyplot as plt 8 9 10 os.environ["CUDA_VISIBLE_DEVICES"] = "1"
1 resize = 224 2 path ="train/" 3 4 def load_data(): 5 imgs = os.listdir(path) 6 num = len(imgs) 7 train_data = np.empty((5000, resize, resize, 3), dtype="int32") 8 train_label = np.empty((5000, ), dtype="int32") 9 test_data = np.empty((5000, resize, resize, 3), dtype="int32") 10 test_label = np.empty((5000, ), dtype="int32") 11 for i in range(5000): 12 if i % 2: 13 train_data[i] = cv2.resize(cv2.imread(path+'/'+ 'dog.' + str(i) + '.jpg'), (resize, resize)) 14 train_label[i] = 1 15 else: 16 train_data[i] = cv2.resize(cv2.imread(path+'/' + 'cat.' + str(i) + '.jpg'), (resize, resize)) 17 train_label[i] = 0 18 for i in range(5000, 10000): 19 if i % 2: 20 test_data[i-5000] = cv2.resize(cv2.imread(path+'/' + 'dog.' + str(i) + '.jpg'), (resize, resize)) 21 test_label[i-5000] = 1 22 else: 23 test_data[i-5000] = cv2.resize(cv2.imread(path+'/' + 'cat.' + str(i) + '.jpg'), (resize, resize)) 24 test_label[i-5000] = 0 25 return train_data, train_label, test_data, test_label
1 def vgg16(): 2 weight_decay = 0.0005 3 nb_epoch = 100 4 batch_size = 32 5 6 # layer1 7 model = keras.Sequential() 8 model.add(layers.Conv2D(64, (3, 3), padding='same', 9 input_shape=(224, 224, 3), kernel_regularizer=regularizers.l2(weight_decay))) 10 model.add(layers.Activation('relu')) 11 model.add(layers.BatchNormalization()) 12 model.add(layers.Dropout(0.3)) 13 # layer2 14 model.add(layers.Conv2D(64, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 15 model.add(layers.Activation('relu')) 16 model.add(layers.BatchNormalization()) 17 model.add(layers.MaxPooling2D(pool_size=(2, 2))) 18 # layer3 19 model.add(layers.Conv2D(128, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 20 model.add(layers.Activation('relu')) 21 model.add(layers.BatchNormalization()) 22 model.add(layers.Dropout(0.4)) 23 # layer4 24 model.add(layers.Conv2D(128, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 25 model.add(layers.Activation('relu')) 26 model.add(layers.BatchNormalization()) 27 model.add(layers.MaxPooling2D(pool_size=(2, 2))) 28 # layer5 29 model.add(layers.Conv2D(256, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 30 model.add(layers.Activation('relu')) 31 model.add(layers.BatchNormalization()) 32 model.add(layers.Dropout(0.4)) 33 # layer6 34 model.add(layers.Conv2D(256, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 35 model.add(layers.Activation('relu')) 36 model.add(layers.BatchNormalization()) 37 model.add(layers.Dropout(0.4)) 38 # layer7 39 model.add(layers.Conv2D(256, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 40 model.add(layers.Activation('relu')) 41 model.add(layers.BatchNormalization()) 42 model.add(layers.MaxPooling2D(pool_size=(2, 2))) 43 # layer8 44 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 45 model.add(layers.Activation('relu')) 46 model.add(layers.BatchNormalization()) 47 model.add(layers.Dropout(0.4)) 48 # layer9 49 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 50 model.add(layers.Activation('relu')) 51 model.add(layers.BatchNormalization()) 52 model.add(layers.Dropout(0.4)) 53 # layer10 54 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 55 model.add(layers.Activation('relu')) 56 model.add(layers.BatchNormalization()) 57 model.add(layers.MaxPooling2D(pool_size=(2, 2))) 58 # layer11 59 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 60 model.add(layers.Activation('relu')) 61 model.add(layers.BatchNormalization()) 62 model.add(layers.Dropout(0.4)) 63 # layer12 64 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 65 model.add(layers.Activation('relu')) 66 model.add(layers.BatchNormalization()) 67 model.add(layers.Dropout(0.4)) 68 # layer13 69 model.add(layers.Conv2D(512, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay))) 70 model.add(layers.Activation('relu')) 71 model.add(layers.BatchNormalization()) 72 model.add(layers.MaxPooling2D(pool_size=(2, 2))) 73 model.add(layers.Dropout(0.5)) 74 # layer14 75 model.add(layers.Flatten()) 76 model.add(layers.Dense(512, kernel_regularizer=regularizers.l2(weight_decay))) 77 model.add(layers.Activation('relu')) 78 model.add(layers.BatchNormalization()) 79 # layer15 80 model.add(layers.Dense(512, kernel_regularizer=regularizers.l2(weight_decay))) 81 model.add(layers.Activation('relu')) 82 model.add(layers.BatchNormalization()) 83 # layer16 84 model.add(layers.Dropout(0.5)) 85 model.add(layers.Dense(2)) 86 model.add(layers.Activation('softmax')) 87 88 return model
1 #if __name__ == '__main__': 2 train_data, train_label, test_data, test_label = load_data() 3 train_data = train_data.astype('float32') 4 test_data = test_data.astype('float32') 5 train_label = keras.utils.to_categorical(train_label, 2) 6 test_label = keras.utils.to_categorical(test_label, 2)
1 #定义训练方法,超参数设置 2 model = vgg16() 3 sgd = tf.keras.optimizers.SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True) #设置优化器为SGD 4 model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy']) 5 history = model.fit(train_data, train_label, 6 batch_size=20, 7 epochs=10, 8 validation_split=0.2, #把训练集中的五分之一作为验证集 9 shuffle=True) 10 scores = model.evaluate(test_data,test_label,verbose=1) 11 print(scores) 12 model.save('model/vgg16dogcat.h5')
1 acc = history.history['accuracy'] # 获取训练集准确性数据 2 val_acc = history.history['val_accuracy'] # 获取验证集准确性数据 3 loss = history.history['loss'] # 获取训练集错误值数据 4 val_loss = history.history['val_loss'] # 获取验证集错误值数据 5 epochs = range(1, len(acc) + 1) 6 plt.plot(epochs, acc, 'bo', label='Trainning acc') # 以epochs为横坐标,以训练集准确性为纵坐标 7 plt.plot(epochs, val_acc, 'b', label='Vaildation acc') # 以epochs为横坐标,以验证集准确性为纵坐标 8 plt.legend() # 绘制图例,即标明图中的线段代表何种含义 9 10 plt.show()
运行结果如下:
标签:layers,decay,VGG,label,add,model,data 来源: https://www.cnblogs.com/-tsir-/p/16275376.html