I originally planned to read the blog to learn, but the writing on Baidu is not clear, so I found that it is better to read the official API, which is very detailed, relatively speaking.
So in the future, I still recommend everyone to read the official documents. Although it may be a bit inconsistent with English, it is actually normal, and the words written by others are also very common English words.
And I also realized that learning Python is a bit biased. Function knowledge is still well written in the official API, which is faster than Baidu's half-day.
conv2d(
input, #input datafilter, #convolution kernel
strides, #step size
padding, #whether padding around 0
use_cudnn_on_gpu=True,
data_format='NHWC',
name=None
)
2-dimensional convolution calculation, given input and filter parameters All are 4-dimensional tensors.
Return:
The same 4-dimensional tensor as the input parameter. . The order between the 4 dimensions is determined by the value of the parameter data_format.
input tensor of shape [ batch , in_height, in_width, in_channels ]
filter tensor of shape [filter_height, filter_width, in_channels , out_channels]
Flattens the filter to a 2-D matrix with shape [filter_height * filter_width * in_channels, output_channels].
For each patch, right-multiplies the filter matrix and the image patch vector
最终返回输出的4维张量: [ batch , out_height, out_width , out_channel ]
strides: 元素长度为4的list,类型为int.Must have strides[0] = strides[3] = 1. strides = [1, stride, stride, 1].
padding: A string from: "SAME", "VALID".
if padding=‘SAME’
out_height=out_width= (in_height=in_width)/stride
else out_height=out_width= ((in_height=in_width)-(filter_height=filter_width)+1)/stride
use_cudnn_on_gpu: An optional bool. Defaults to True.
data_format: An optional string from: "NHWC", "NCHW". With the default format "NHWC", the data is stored in the order of: [batch, height, width, channels]. Alternatively, the format could be "NCHW", the data storage order of: [batch, channels, height, width].
name: A name for the operation (optional).
max_pool(
value,
ksize,
strides,
padding,
data_format='NHWC',
name=None
)
performs max pooling calculation on input data
Returns:
A Tensor of format specified by data_format. The max pooled output tensor.
value: A 4-D Tensor of the format specified by data_format.
ksize: a list with an element length of 4, the type is int. Indicates the size of the pooling kernel
strides: a list with an element length of 4, the type is int. Indicates the pooling step size
padding: A string , either 'VALID' or 'SAME'. As defined above
data_format: A string. 'NHWC', 'NCHW' and 'NCHW_VECT_C' are supported.
name: Optional name for the operation.
Must have strides = [1, stride, stride, 1].
ksizes = [1, ksize, ksize, 1].