matlab中sconv和conv的区别,基于Keras中Conv1D和Conv2D的区别说明

如有错误&＃xff0c;欢迎斧正。

我的答案是&＃xff0c;在Conv2D输入通道为1的情况下&＃xff0c;二者是没有区别或者说是可以相互转化的。首先&＃xff0c;二者调用的最后的代码都是后端代码(以TensorFlow为例&＃xff0c;在tensorflow_backend.py里面可以找到)&＃xff1a;

x &＃61; tf.nn.convolution(

input&＃61;x,

filter&＃61;kernel,

dilation_rate&＃61;(dilation_rate,),

strides&＃61;(strides,),

padding&＃61;padding,

data_format&＃61;tf_data_format)

区别在于input和filter传递的参数不同&＃xff0c;input不必说&＃xff0c;filter&＃61;kernel是什么呢&＃xff1f;

我们进入Conv1D和Conv2D的源代码看一下。他们的代码位于layers/convolutional.py里面&＃xff0c;二者继承的都是基类_Conv(Layer)。

进入_Conv类查看代码可以发觉以下代码&＃xff1a;

self.kernel_size &＃61; conv_utils.normalize_tuple(kernel_size, rank, &＃39;kernel_size&＃39;)

……#中间代码省略

input_dim &＃61; input_shape[channel_axis]

kernel_shape &＃61; self.kernel_size &＃43; (input_dim, self.filters)

我们假设&＃xff0c;Conv1D的input的大小是(600,300)&＃xff0c;而Conv2D的input大小是(m,n,1)&＃xff0c;二者kernel_size为3。

进入conv_utils.normalize_tuple函数可以看到&＃xff1a;

def normalize_tuple(value, n, name):

"""Transforms a single int or iterable of ints into an int tuple.

# Arguments

value: The value to validate and convert. Could an int, or any iterable

of ints.

n: The size of the tuple to be returned.

name: The name of the argument being validated, e.g. "strides" or

"kernel_size". This is only used to format error messages.

# Returns

A tuple of n integers.

# Raises

ValueError: If something else than an int/long or iterable thereof was

passed.

"""

if isinstance(value, int):

return (value,) * n

else:

try:

value_tuple &＃61; tuple(value)

except TypeError:

raise ValueError(&＃39;The &＃96;&＃39; &＃43; name &＃43; &＃39;&＃96; argument must be a tuple of &＃39; &＃43;

str(n) &＃43; &＃39; integers. Received: &＃39; &＃43; str(value))

if len(value_tuple) !&＃61; n:

raise ValueError(&＃39;The &＃96;&＃39; &＃43; name &＃43; &＃39;&＃96; argument must be a tuple of &＃39; &＃43;

str(n) &＃43; &＃39; integers. Received: &＃39; &＃43; str(value))

for single_value in value_tuple:

try:

int(single_value)

except ValueError:

raise ValueError(&＃39;The &＃96;&＃39; &＃43; name &＃43; &＃39;&＃96; argument must be a tuple of &＃39; &＃43;

str(n) &＃43; &＃39; integers. Received: &＃39; &＃43; str(value) &＃43; &＃39; &＃39;

&＃39;including element &＃39; &＃43; str(single_value) &＃43; &＃39; of type&＃39; &＃43;

&＃39; &＃39; &＃43; str(type(single_value)))

return value_tuple

所以上述代码得到的kernel_size是kernel的实际大小&＃xff0c;根据rank进行计算&＃xff0c;Conv1D的rank为1&＃xff0c;Conv2D的rank为2&＃xff0c;如果是Conv1D&＃xff0c;那么得到的kernel_size就是(3,)如果是Conv2D&＃xff0c;那么得到的是(3,3)

input_dim &＃61; input_shape[channel_axis] kernel_shape &＃61; self.kernel_size &＃43; (input_dim, self.filters)

又因为以上的inputdim是最后一维大小(Conv1D中为300&＃xff0c;Conv2D中为1)&＃xff0c;filter数目我们假设二者都是64个卷积核。

因此&＃xff0c;Conv1D的kernel的shape实际为&＃xff1a;

(3,300,64)

而Conv2D的kernel的shape实际为&＃xff1a;