深度学习实战篇之 ( 十三) -- TensorFlow之AlexNet

def inference(images, batch_size, n_classes,drop_rate):
    # 卷积层1：卷积核大小为11*11 卷积核个数：96,步长：4
    conv1 = Conv_layer(names = 'conv1_scope', input = images , w_shape = [11, 11, 3, 96], b_shape = [96], strid = [4, 4], is_padding = False)
    print("---------conv1:{}".format(conv1))
    # 最大池化1：采样区域：3*3, 步长：默认为1
    down_sample1 = Avg_pool_lrn(names = 'avg_pooling1', input = conv1 , ksize = [1, 2, 2, 1], is_lrn = False)
    print("---------down_sample1:{}".format(down_sample1))
    # 卷积层2：积核大小为5*5 ,卷积核个数：256, 步长：1
    conv2 = Conv_layer(names = 'conv2_scope', input = down_sample1 , w_shape = [5, 5, 96, 256], b_shape = [256], strid = [1, 1], is_padding = False)
    # 最大池化2：采样区域：3*3, 步长：默认为1
    down_sample2 = Avg_pool_lrn(names = 'avg_pooling2', input = conv2 , ksize = [1, 3, 3, 1], is_lrn = False)
    # 卷积层3：积核大小为3*3 ,卷积核个数：384, 步长：1，采用padding。
    conv3 = Conv_layer(names = 'conv3_scope', input = down_sample2 , w_shape = [3, 3, 256, 384], b_shape = [384], strid = [1, 1], is_padding = True)
    # 卷积层4：积核大小为3*3 ,卷积核个数：384, 步长：1，采用padding。
    conv4 = Conv_layer(names = 'conv4_scope', input = conv3 , w_shape = [3, 3, 384, 384], b_shape = [384], strid = [1, 1], is_padding = True)
    # 卷积层5：积核大小为3*3 ,卷积核个数：256, 步长：1，采用padding。（在理论篇文章中遗漏了这一层，切记，注意）
    conv5 = Conv_layer(names = 'conv4_scope', input = conv3 , w_shape = [3, 3, 384, 256], b_shape = [256], strid = [1, 1], is_padding = True)
    reshape = tf.reshape(conv3, shape=[batch_size, -1])
    dim = reshape.get_shape()[1].value
    # 全连接层1：输出84
    local_1 = local_layer(names = 'local1_scope', input = reshape , w_shape = [dim, 84], b_shape = [84])
     # 全连接层2：输出5
    local_2 = local_layer(names = 'local2_scope', input = local_1 , w_shape = [84, 5], b_shape = [5])
    print(local_2.shape)

    return local_2