ShuffleNetV1网络结构训练模型（图像分类）

文章目录

• • 1.ShuffleNetV1文章
  • 2.重新搭建ShuffleNetV1模型结构（便于训练）
  • • （1）关于数据集和训练的整个代码可以参考我这篇文章
    • （2）改换代码中的网络结构
  • 3.训练结果
  • 4.图像测试

1.ShuffleNetV1文章

https://mydreamambitious.blog.csdn.net/article/details/124675932

---

2.重新搭建ShuffleNetV1模型结构（便于训练） （1）关于数据集和训练的整个代码可以参考我这篇文章

https://mydreamambitious.blog.csdn.net/article/details/123966676

（2）改换代码中的网络结构

注：只要将代码中的网络结构换成下面的这个ShuffleNetV1结构即可（如果采用的是面向对象的方式设计的网络结构的话，在训练数据的时候需要做一些调整）并且再将代码中的一些名称修改一下即可。
需要修改的名称如下：

# 通道重排
def Shuffle_Channels(inputs, groups):
    # 获取图片batch，高度，宽度，通道数
    batch, h, w, c = inputs.shape
    # 对通道进行变形
    input_reshape = tf.reshape(inputs, [-1, h, w, groups, c // groups])
    # 退通道进行转置
    input_transposed = tf.transpose(input_reshape, [0, 1, 2, 4, 3])
    # 再对通道进行变形
    output_reshape = tf.reshape(input_transposed, [-1, h, w, c])
    return output_reshape

#分组卷积
def group_conv(inputs,filters,kernel,strides,groups):
    #对通道进行切分
    conv_num_layers=tf.split(inputs,groups,axis=3)
    conv_side_layers=[]
    for layer in conv_num_layers:
        conv_side_layers.append(
            layers.Conv2D(filters//groups,kernel_size=kernel,strides=strides,padding='same')(layer)
        )
    x_out=tf.concat(
        conv_side_layers,axis=-1
    )
    return x_out

def ShuffleNetUnitB(inputs, filter, groups):
    x=group_conv(inputs,filters=filter,kernel=[1,1],strides=[1,1],groups=groups)
    # x = layers.Conv2D(filter // 4, kernel_size=[1, 1], strides=[1, 1], padding='same')(inputs)
    x = layers.BatchNormalization()(x)
    x = layers.Activation('relu')(x)

    x = Shuffle_Channels(x, groups)

    x = layers.DepthwiseConv2D(kernel_size=[3, 3], strides=[1, 1], padding='same')(x)
    x = layers.BatchNormalization()(x)

    x = group_conv(x, filters=filter, kernel=[1, 1], strides=[1, 1], groups=groups)
    # x = layers.Conv2D(filter, kernel_size=[1, 1], strides=[1, 1], groups=groups, padding='same')(x)
    x = layers.BatchNormalization()(x)

    x_out = tf.add(inputs, x)
    x_out = layers.Activation('relu')(x_out)

    return x_out


def ShuffleNetUnitC(inputs, filter, inputFilter, groups):
    outputChannels = filter - inputFilter

    x = group_conv(inputs, filters=filter, kernel=[1, 1], strides=[1, 1], groups=groups)
    # x = layers.Conv2D(filter // 4, kernel_size=[1, 1], strides=[1, 1], padding='same')(inputs)
    x = layers.BatchNormalization()(x)
    x = layers.Activation('relu')(x)

    x = Shuffle_Channels(x, groups)

    x = layers.DepthwiseConv2D(kernel_size=[3, 3], strides=[2, 2], padding='same')(x)
    x = layers.BatchNormalization()(x)

    x = group_conv(x, filters=outputChannels, kernel=[1, 1], strides=[1, 1], groups=groups)
    # x = layers.Conv2D(outputChannels, kernel_size=[1, 1], strides=[1, 1], padding='same')(x)
    x = layers.BatchNormalization()(x)

    x_avgpool3x3 = layers.AveragePooling2D(pool_size=[3, 3], strides=[2, 2], padding='same')(inputs)
    # 对channel进行堆叠
    x_out = tf.concat([x, x_avgpool3x3],
                      axis=3)
    x_out = layers.Activation('relu')(x_out)

    return x_out


def ShuffleNetV1Model(numlayers, stage2Filter, stage3Filter, stage4Filter, groups,input_shape=(224,224,3)):
    input = keras.Input(input_shape)

    x = layers.Conv2D(24, kernel_size=[3, 3], strides=[2, 2], padding='same')(input)
    x = layers.BatchNormalization()(x)
    x = layers.Activation('relu')(x)

    x = layers.MaxPool2D(pool_size=[3, 3], strides=[2, 2], padding='same')(x)

    def StageX(inputs, numlayers, filter, inputFilter,groups):
        x_out=ShuffleNetUnitC(inputs, filter, inputFilter, groups)
        for i in range(numlayers):
            x_out=ShuffleNetUnitB(x_out, filter, groups)
        return x_out

    x = StageX(x, numlayers[0], stage2Filter, inputFilter=24, groups=groups)
    x = StageX(x, numlayers[1], stage3Filter, inputFilter=240, groups=groups)
    x = StageX(x, numlayers[2], stage4Filter, inputFilter=480, groups=groups)

    x = layers.GlobalAveragePooling2D()(x)
    x = layers.Dense(2)(x)
    x_out = layers.Activation('softmax')(x)

    model = Model(inputs=input, outputs=x_out)
    return model


model_shufflenetv1 = ShuffleNetV1Model(numlayers=[3, 7, 3], stage2Filter=240, stage3Filter=480,
                                  stage4Filter=960, groups=3,input_shape=(224, 224, 3))
model_shufflenetv1.summary()

……

---

3.训练结果

由于数据集不是很多，只有近1000张图片，并且之包含两个类别（猫-cat,狗-dot）所以训练的最后效果并不怎么样（中间还需要调参），读者可以根据自己的需要收集图片进行训练。

4.图像测试

只需要将flask中加载的权重文件换成自己训练的就可以了（flask这部分代码在从Github下载的文件中）。
https://github.com/KeepTryingTo/-.git

flask.py文件：


前端文件：index.html