【深度学习】使用tensorflow实现AlexNet

转载注明出处：http://blog.csdn.net/accepthjp/article/details/69999309

AlexNet是2012年ImageNet比赛的冠军，虽然过去了很长时间，但是作为深度学习中的经典模型，AlexNet不但有助于我们理解其中所使用的很多技巧，而且非常有助于提升我们使用深度学习工具箱的熟练度。尤其是我刚入门深度学习，迫切需要一个能让自己熟悉tensorflow的小练习，于是就有了这个小玩意儿......

先放上我的代码：https://github.com/hjptriplebee/AlexNet_with_tensorflow

如果想运行代码，详细的配置要求都在上面链接的readme文件中了。本文建立在一定的tensorflow基础上，不会对太细的点进行说明。

模型结构

关于模型结构网上的文献很多，我这里不赘述，一会儿都在代码里解释。

有一点需要注意，AlexNet将网络分成了上下两个部分，在论文中两部分结构完全相同，唯一不同的是他们放在不同GPU上训练，因为每一层的feature map之间都是独立的（除了全连接层），所以这相当于是提升训练速度的一种方法。很多AlexNet的复现都将上下两部分合并了，因为他们都是在单个GPU上运行的。虽然我也是在单个GPU上运行，但是我还是很想将最原始的网络结构还原出来，所以我的代码里也是分开的。

模型定义

def maxPoolLayer(x, kHeight, kWidth, strideX, strideY, name, padding = "SAME"):    """max-pooling"""    return tf.nn.max_pool(x, ksize = [1, kHeight, kWidth, 1],                          strides = [1, strideX, strideY, 1], padding = padding, name = name)def dropout(x, keepPro, name = None):    """dropout"""    return tf.nn.dropout(x, keepPro, name)def LRN(x, R, alpha, beta, name = None, bias = 1.0):    """LRN"""    return tf.nn.local_response_normalization(x, depth_radius = R, alpha = alpha,                                              beta = beta, bias = bias, name = name)def fcLayer(x, inputD, outputD, reluFlag, name):    """fully-connect"""    with tf.variable_scope(name) as scope:        w = tf.get_variable("w", shape = [inputD, outputD], dtype = "float")        b = tf.get_variable("b", [outputD], dtype = "float")        out = tf.nn.xw_plus_b(x, w, b, name = scope.name)        if reluFlag:            return tf.nn.relu(out)        else:            return outdef convLayer(x, kHeight, kWidth, strideX, strideY,              featureNum, name, padding = "SAME", groups = 1):#group为2时等于AlexNet中分上下两部分    """convlutional"""    channel = int(x.get_shape()[-1])#获取channel    conv = lambda a, b: tf.nn.conv2d(a, b, strides = [1, strideY, strideX, 1], padding = padding)#定义卷积的匿名函数    with tf.variable_scope(name) as scope:        w = tf.get_variable("w", shape = [kHeight, kWidth, channel/groups, featureNum])        b = tf.get_variable("b", shape = [featureNum])        xNew = tf.split(value = x, num_or_size_splits = groups, axis = 3)#划分后的输入和权重        wNew = tf.split(value = w, num_or_size_splits = groups, axis = 3)        featureMap = [conv(t1, t2) for t1, t2 in zip(xNew, wNew)] #分别提取feature map        mergeFeatureMap = tf.concat(axis = 3, values = featureMap) #feature map整合        # print mergeFeatureMap.shape        out = tf.nn.bias_add(mergeFeatureMap, b)        return tf.nn.relu(tf.reshape(out, mergeFeatureMap.get_shape().as_list()), name = scope.name) #relu后的结果

定义了卷积、pooling、LRN、dropout、全连接五个模块，其中卷积模块因为将网络的上下两部分分开了，所以比较复杂。接下来定义AlexNet。

class alexNet(object):    """alexNet model"""    def __init__(self, x, keepPro, classNum, skip, modelPath = "bvlc_alexnet.npy"):        self.X = x        self.KEEPPRO = keepPro        self.CLASSNUM = classNum        self.SKIP = skip        self.MODELPATH = modelPath        #build CNN        self.buildCNN()    def buildCNN(self):        """build model"""        conv1 = convLayer(self.X, 11, 11, 4, 4, 96, "conv1", "VALID")        pool1 = maxPoolLayer(conv1, 3, 3, 2, 2, "pool1", "VALID")        lrn1 = LRN(pool1, 2, 2e-05, 0.75, "norm1")        conv2 = convLayer(lrn1, 5, 5, 1, 1, 256, "conv2", groups = 2)        pool2 = maxPoolLayer(conv2, 3, 3, 2, 2, "pool2", "VALID")        lrn2 = LRN(pool2, 2, 2e-05, 0.75, "lrn2")        conv3 = convLayer(lrn2, 3, 3, 1, 1, 384, "conv3")        conv4 = convLayer(conv3, 3, 3, 1, 1, 384, "conv4", groups = 2)        conv5 = convLayer(conv4, 3, 3, 1, 1, 256, "conv5", groups = 2)        pool5 = maxPoolLayer(conv5, 3, 3, 2, 2, "pool5", "VALID")        fcIn = tf.reshape(pool5, [-1, 256 * 6 * 6])        fc1 = fcLayer(fcIn, 256 * 6 * 6, 4096, True, "fc6")        dropout1 = dropout(fc1, self.KEEPPRO)        fc2 = fcLayer(dropout1, 4096, 4096, True, "fc7")        dropout2 = dropout(fc2, self.KEEPPRO)        self.fc3 = fcLayer(dropout2, 4096, self.CLASSNUM, True, "fc8")    def loadModel(self, sess):        """load model"""        wDict = np.load(self.MODELPATH, encoding = "bytes").item()        #for layers in model        for name in wDict:            if name not in self.SKIP:                with tf.variable_scope(name, reuse = True):                    for p in wDict[name]:                        if len(p.shape) == 1:                             #bias 只有一维                            sess.run(tf.get_variable('b', trainable = False).assign(p))                        else:                            #weights                             sess.run(tf.get_variable('w', trainable = False).assign(p))

buildCNN函数完全按照alexnet的结构搭建网络。

loadModel函数从模型文件中读取参数，采用的模型文件见github上的readme说明。
至此，我们定义了完整的模型，下面开始测试模型。

模型测试

ImageNet训练的AlexNet有很多类，几乎包含所有常见的物体，因此我们随便从网上找几张图片测试。比如我直接用了之前做项目的渣土车图片：

然后编写测试代码：

#some paramsdropoutPro = 1classNum = 1000skip = []#get testImagetestPath = "testModel"testImg = []for f in os.listdir(testPath):    testImg.append(cv2.imread(testPath + "/" + f))imgMean = np.array([104, 117, 124], np.float)x = tf.placeholder("float", [1, 227, 227, 3])model = alexnet.alexNet(x, dropoutPro, classNum, skip)score = model.fc3softmax = tf.nn.softmax(score)with tf.Session() as sess:    sess.run(tf.global_variables_initializer())    model.loadModel(sess) #加载模型    for i, img in enumerate(testImg):        #img preprocess        test = cv2.resize(img.astype(np.float), (227, 227)) #resize成网络输入大小        test -= imgMean #去均值        test = test.reshape((1, 227, 227, 3)) #拉成tensor        maxx = np.argmax(sess.run(softmax, feed_dict = {x: test}))        res = caffe_classes.class_names[maxx] #取概率最大类的下标        #print(res)        font = cv2.FONT_HERSHEY_SIMPLEX        cv2.putText(img, res, (int(img.shape[0]/3), int(img.shape[1]/3)), font, 1, (0, 255, 0), 2)#绘制类的名字        cv2.imshow("demo", img)         cv2.waitKey(5000) #显示5秒

如上代码所示，首先需要设置一些参数，然后读取指定路径下的测试图像，再对模型做一个初始化，最后是真正测试代码。测试结果如下：