import keras
from keras.models import Sequential
from keras.layers import Dense,MaxPooling2D,Input,Flatten,Convolution2D,Dropout
from keras.optimizers import SGD
from keras.callbacks import TensorBoard,ModelCheckpoint
from PIL import Image
import os
import numpy  as np
from scipy import misc
root_path = os.getcwd()


def load_data():
    tran_imags = []
    labels = []
    seq_names = ['cat','dog']
    for seq_name in seq_names:
        frames = sorted(os.listdir(os.path.join(root_path,'data','train_data', seq_name)))
        for frame in frames:
            imgs = [os.path.join(root_path, 'data', 'train_data', seq_name, frame)]
            imgs = np.array(Image.open(imgs[0]))
            tran_imags.append(imgs)
            if seq_name=='cat':
                labels.append(0)
            else:
                labels.append(1)
    return np.array(tran_imags), np.array(labels)


def vgg_cat_dog(weights_path=None):
    model = Sequential()
    model.add(Convolution2D(64,(3,3),activation='relu',padding='same',input_shape=(224,224,3)))
    model.add(Convolution2D(64,(3,3),activation='relu',padding='same'))
    model.add(MaxPooling2D((2,2),strides=(2,2)))

    model.add(Convolution2D(128,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(128,(3,3),activation='relu',padding='same'))
    model.add(MaxPooling2D((2,2),strides=(2,2)))

    model.add(Convolution2D(256,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(256,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(256,(3,3),activation='relu',padding='same'))
    model.add(MaxPooling2D((2,2),strides=(2,2)))

    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(MaxPooling2D((2,2),strides=(2,2)))

    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(Convolution2D(512,(3,3),activation='relu',padding='same'))
    model.add(MaxPooling2D((2,2),strides=(2,2)))


    if weights_path:
        model.load_weights(weights_path)

    model.add(Flatten())
    model.add(Dense(500,activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(100,activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(2,activation='softmax'))
    return model


train_data,train_labs = load_data()
train = True
if train:
 model = vgg_cat_dog('model/vgg_16_without_fc.h5')
 sgd = SGD(lr=1e-6,momentum=0.9,decay=1e-6,nesterov=True)
 model.compile(optimizer=sgd,loss='categorical_crossentropy',metrics=['accuracy'])
 tensorboard = TensorBoard(log_dir='log/cat_dog',histogram_freq=10,write_images=True)
 ckpt = ModelCheckpoint(os.path.join(root_path,'model','cat_dog','vgg.h5'),verbose=1,period=5)
 model.fit(train_data,keras.utils.to_categorical(train_labs),batch_size=32,epochs=50,verbose=1,
          callbacks=[tensorboard,ckpt],validation_split=0.1,shuffle=True)
else:
 model = vgg_cat_dog()
 model.load_weights('model/cat_dog/vgg.h5')
 frames = sorted(os.listdir(os.path.join(root_path,'data','test', 'cat')))
 for frame in frames:
    imgs = [os.path.join(root_path, 'data', 'test', 'cat', frame)]
    im = np.expand_dims(np.array(misc.imresize(Image.open(imgs[0]), [224, 224])), axis=0)
    cal = np.argmax(model.predict(im))
    if cal:
        print('dog')
    else:
        print('cat')
 

使用在imagenet上的預訓練好的前13層的權重進行fine-tune，訓練集貓狗各420張左右，進行了50輪訓練，效果還不錯

記：1,開始未加dropout.   微調時，引數調的不多，第一次設定的學習率過大（0.1），損失居高不下一條直線，精度也是一條直線（0.1）。 第二次設定的1e-8，網路開始收斂，比較慢，最後選擇1e-6,訓練集精度達到95左右，驗證集只有50左右的精度，過擬合。

2. 新增dropout層，引數不變，驗證集精度飆升，由上圖可見40輪左右基本穩定下來了，不得不感嘆，dropout真是好啊。

附：keras下：1.vgg16帶全連線層權重帶全連線連結     2.vgg16不帶全連線權重