Two dimensional convolution -- use of torch.nn.conv2d

First open the official function , see conv2d How to use .

conv The function consists of three required parameters ,

        in_channels Refers to the number of input channels ,out_channels Refers to the number of output channels ,kernel_size Is the number of cores of the convolution kernel ,

There are also some non mandatory parameters . for example padding and stride Parameters , The detailed functions of these two parameters are shown in the following figure .

stride The control is to convolute the rows and columns of several matrices at a time ,padding Is to fill in the data to be processed .

1.padding=0、stride=0 The situation of .

2.padding=2、stride=0 The situation of .

https://github.com/vdumoulin/conv_arithmetic/blob/master/gif/arbitrary_padding_no_strides.gif

3.padding=1、stride=0 The situation of .

conv_arithmetic/same_padding_no_strides.gif at master · vdumoulin/conv_arithmetic · GitHub

about 2 More detailed demonstration of dimensional convolution github link ：

conv_arithmetic/README.md at master · vdumoulin/conv_arithmetic · GitHub

adopt debug Code , We can see that the number of channels of our pictures is determined by 3 The passage was turned into 6 passageway , But at the end add tensorboard In the time , Need to be converted to 3 passageway , Therefore use torch.reshape convert .

#coding=utf-8
# Main learning torch.nn.conv2d()
import torch
import torchvision
from torch.utils.data import DataLoader
from PIL import Image
from torch.utils.tensorboard import SummaryWriter

dataset = torchvision.datasets.CIFAR10("./dataset",train=False,transform=torchvision.transforms.ToTensor(),download=True)
dataloader = DataLoader(dataset,batch_size=64)

class this_network(torch.nn.Module):
    def __init__(self):
        super(this_network,self).__init__()
        self.conv1 = torch.nn.Conv2d(in_channels=3,out_channels=6,kernel_size=3,stride=1,padding=0)
    
    def forward(self,x):
        x = self.conv1(x)
        return x

init_network = this_network()
print(init_network)
writer = SummaryWriter("./logs")

step = 0
for data in dataloader:
    imgs,targets = data
    output = init_network(imgs)
    #print(imgs.shape)
    #print(output.shape)

    # Use tensorboard Data browsing 
    writer.add_images("input",imgs,step)
    output = torch.reshape(output,(-1,3,30,30))
    writer.add_images("output",output,step)
    step = step + 1
    

design sketch