标签：conv pooling channels tf shape cnn tensorflow out 3d

def CNN_3d(x, out_channels_0, out_channels_1, add_relu=True):
    '''Add a 3d convlution layer with relu and max pooling layer.

    Args:
        x: a tensor with shape [batch, in_depth, in_height, in_width, in_channels]
        out_channels: a number
        filter_size: a number
        pooling_size: a number

    Returns:
        a flattened tensor with shape [batch, num_features]

    Raises:
    '''
    in_channels = x.shape[-1]
    weights_0 = tf.get_variable(
        name='filter_0',
        shape=[3, 3, 3, in_channels, out_channels_0],
        dtype=tf.float32,
        initializer=tf.random_uniform_initializer(-0.01, 0.01))
    bias_0 = tf.get_variable(
        name='bias_0',
        shape=[out_channels_0],
        dtype=tf.float32,
        initializer=tf.zeros_initializer())

    conv_0 = tf.nn.conv3d(x, weights_0, strides=[1, 1, 1, 1, 1], padding="SAME")
    print('conv_0 shape: %s' %conv_0.shape)
    conv_0 = conv_0 + bias_0

    if add_relu:
        conv_0 = tf.nn.elu(conv_0)

    pooling_0 = tf.nn.max_pool3d(
        conv_0, 
        ksize=[1, 3, 3, 3, 1],
        strides=[1, 3, 3, 3, 1], 
        padding="SAME")
    print('pooling_0 shape: %s' %pooling_0.shape)

    #layer_1
    weights_1 = tf.get_variable(
        name='filter_1',
        shape=[3, 3, 3, out_channels_0, out_channels_1],
        dtype=tf.float32,
        initializer=tf.random_uniform_initializer(-0.01, 0.01))
    bias_1 = tf.get_variable(
        name='bias_1',
        shape=[out_channels_1],
        dtype=tf.float32,
        initializer=tf.zeros_initializer())

    conv_1 = tf.nn.conv3d(pooling_0, weights_1, strides=[1, 1, 1, 1, 1], padding="SAME")
    print('conv_1 shape: %s' %conv_1.shape)
    conv_1 = conv_1 + bias_1

    if add_relu:
        conv_1 = tf.nn.elu(conv_1)

    pooling_1 = tf.nn.max_pool3d(
        conv_1, 
        ksize=[1, 3, 3, 3, 1],
        strides=[1, 3, 3, 3, 1], 
        padding="SAME")
    print('pooling_1 shape: %s' %pooling_1.shape)

    return tf.contrib.layers.flatten(pooling_1)

调用比如CNN_3d(x, 32, 16)
其中输入x是一个类似[batch_size,9,50,50,12]
则输出是[batch_size,576]

标签：conv,pooling,channels,tf,shape,cnn,tensorflow,out,3d
来源： https://blog.csdn.net/guotong1988/article/details/100728970

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。