一、簡介

VGG網路在2014年的 ILSVRC localization and classification 兩個問題上分別取得了第一名和第二名。VGG網路非常深，通常有16－19層，如果自己訓練網路模型的話很浪費時間和計算資源。因此這裡採用一種方法獲取VGG19模型的模型資料，從而能夠更快速的應用到自己的任務中來，

本文在載入模型資料的同時，還視覺化圖片在網路傳播過程中，每一層的輸出特徵圖。讓我們能夠更直接的觀察網路傳播的狀況。

執行環境為spyder，Python3.5，tensorflow1.2.1  模型名稱為： imagenet-vgg-verydeep-19.mat 大家可以在網上下載。

二、VGG19模型結構

模型的每一層結構如下圖所示： 

三、程式碼

    #載入VGG19模型並可視化一張圖片前向傳播的過程中每一層的輸出
    #引入包
    import tensorflow as tf
    import numpy as np
    import matplotlib.pyplot as plt
    import scipy.io
    import scipy.misc
    #定義一些函式
    #卷積
    def _conv_layer(input, weights, bias):
           conv = tf.nn.conv2d(input, tf.constant(weights), strides=(1, 1, 1, 1),
                   padding='SAME')
           return tf.nn.bias_add(conv, bias)
    #池化
    def _pool_layer(input):
           return tf.nn.max_pool(input, ksize=(1, 2, 2, 1), strides=(1, 2, 2, 1),
                   padding='SAME')
    #減畫素均值操作
    def preprocess(image, mean_pixel):
           return image - mean_pixel
    #加畫素均值操作
    def unprocess(image, mean_pixel):
           return image + mean_pixel
    #讀
    def imread(path):
           return scipy.misc.imread(path).astype(np.float)
    #儲存
    def imsave(path, img):
           img = np.clip(img, 0, 255).astype(np.uint8)
           scipy.misc.imsave(path, img)
    print ("Functions for VGG ready")
    #定義VGG的網路結構，用來儲存網路的權重和偏置引數
    def net(data_path, input_image):
            #拿到每一層對應的引數
            layers = (
                'conv1_1', 'relu1_1', 'conv1_2', 'relu1_2', 'pool1',
                'conv2_1', 'relu2_1', 'conv2_2', 'relu2_2', 'pool2',
                'conv3_1', 'relu3_1', 'conv3_2', 'relu3_2', 'conv3_3',
                'relu3_3', 'conv3_4', 'relu3_4', 'pool3',
                'conv4_1', 'relu4_1', 'conv4_2', 'relu4_2', 'conv4_3',
                'relu4_3', 'conv4_4', 'relu4_4', 'pool4',
                'conv5_1', 'relu5_1', 'conv5_2', 'relu5_2', 'conv5_3',
                'relu5_3', 'conv5_4', 'relu5_4'
            )
            data = scipy.io.loadmat(data_path)
            #原網路在訓練的過程中，對每張圖片三通道都執行了減均值的操作，這裡也要減去均值
            mean = data['normalization'][0][0][0]
            mean_pixel = np.mean(mean, axis=(0, 1))
            #print(mean_pixel)
            #取到權重引數W和b,這裡運氣好的話，可以查到VGG模型中每層的引數含義，查不到的
            #話可以打印出weights，然後列印每一層的shape，推出其中每一層代表的含義
            weights = data['layers'][0]
            #print(weights)
            net = {}
            current = input_image
            #取到w和b
            for i, name in enumerate(layers):
                #:4的含義是隻看每一層的前三個字母，從而進行判斷
                kind = name[:4]
                if kind == 'conv':
                    kernels, bias = weights[i][0][0][0][0]
                    # matconvnet: weights are [width, height, in_channels, out_channels]\n",
                    # tensorflow: weights are [height, width, in_channels, out_channels]\n",
                    #這裡width和height是顛倒的，所以要做一次轉置運算
                    kernels = np.transpose(kernels, (1, 0, 2, 3))
                    #將bias轉換為一個維度
                    bias = bias.reshape(-1)
                    current = _conv_layer(current, kernels, bias)
                elif kind == 'relu':
                    current = tf.nn.relu(current)
                elif kind == 'pool':
                    current = _pool_layer(current)
                net[name] = current
            assert len(net) == len(layers)
            return net, mean_pixel, layers
    print ("Network for VGG ready")
    #cwd  = os.getcwd()
    #這裡用的是絕對路徑
    VGG_PATH = "F:/mnist/imagenet-vgg-verydeep-19.mat"
    #需要視覺化的圖片路徑，這裡是一隻小貓
    IMG_PATH = "D:/VS2015Program/cat.jpg"
    input_image = imread(IMG_PATH)
    #獲取影象shape
    shape = (1,input_image.shape[0],input_image.shape[1],input_image.shape[2]) 
    #開始會話
    with tf.Session() as sess:
            image = tf.placeholder('float', shape=shape)
            #呼叫net函式
            nets, mean_pixel, all_layers = net(VGG_PATH, image)
            #減均值操作（由於VGG網路圖片傳入前都做了減均值操作，所以這裡也用相同的預處理
            input_image_pre = np.array([preprocess(input_image, mean_pixel)])
            layers = all_layers # For all layers \n",
            # layers = ('relu2_1', 'relu3_1', 'relu4_1')\n",
            for i, layer in enumerate(layers):
                print ("[%d/%d] %s" % (i+1, len(layers), layer))
                features = nets[layer].eval(feed_dict={image: input_image_pre})
                print (" Type of 'features' is ", type(features))
                print (" Shape of 'features' is %s" % (features.shape,))
                # Plot response \n",
                #畫出每一層
                if 1:
                    plt.figure(i+1, figsize=(10, 5))
                    plt.matshow(features[0, :, :, 0], cmap=plt.cm.gray, fignum=i+1)
                    plt.title("" + layer)
                    plt.colorbar()
                    plt.show()
 

四、程式執行結果

1、print(weights)的結果： 

2、程式執行最終結果：  中間層數太多，這裡就不展示了。程式最後兩層的視覺化結果：