Tensorflow tensor convolution, input and convolution kernel dimension understanding

import tensorflow as tf


inputValue = tf.constant([
    #1 A tensor 
    [
        #3 That's ok 3 Column 2 depth 
        [[2, 5], [3, 3], [8, 2]],
        [[6, 1], [1, 2], [5, 4]],
        [[7, 9], [2, 3], [-1, 3]]
    ]
])

kernels = tf.constant([
    # 2 That's ok 
    [
        # 2 Column 
        [
            # 2 depth 
            [3, 1, -3], [1, -1, 7]
        ],
        [
            [-2, 2, -5], [2, 7, 3]
        ]
    ],
    [
        # 2 Column 
        [
            [-1, 3, 1], [-3, -8, 6]
        ],
        [
            [4, 6, 8], [5, 9, -5]
        ]
    ]
])

validResult = tf.nn.conv2d(inputValue, kernels, [1, 1, 1, 1], "VALID")

print(validResult)

The contents of each tensor are represented in the red circle , The square brackets circled by the blue circle indicate the dimension of number .

The brackets circled in blue indicate the dimension of row

Red indicates the dimension of column

The two numbers in the red circle column , Indicates that the depth is 2

tf in ,shape This shows 1 individual 3 That's ok 3 Column 2 Tensor of depth , stay cnn In training ,minibatch There are multiple inputs , It's just “1” This Number Add

The dimension of convolution kernel

The dimension of convolution kernel (2,2,2,3), Express 3 individual 2 That's ok 2 Column 2 Convolution kernel of depth （ That's ok , Column , depth , Number ）

Red brackets represent lines 、 Blue represents column 、 Yellow represents depth 、 Three numbers in the depth represent numbers

validResult = tf.nn.conv2d(inputValue, kernels, [1, 1, 1, 1], "VALID")

“VALID” Express padding The pattern is valid,[1,1,1,1] It means in number 、 That's ok 、 Column 、 Depth in these four dimensions stride Namely 1、1、1、1