torch Great God, please ignore this article ...

1, A simple example review DataSet

from torch.utils.data import Dataset

class dataset(Dataset):
    def __init__(self):
        #  We need to convert to array, Otherwise, the running result will be very strange 
        self.data = np.array([[1,1,1,1],
                [2,2,2,2],
                [3,3,3,3],
                [4,4,4,4],
                [5,5,5,5],
                [6,6,6,6]])        
                
    def __len__(self):
        return len(self.data)
    
    def __getitem__(self, idx):
        # return data[idx], This sentence is missing self, Report errors ...
        return self.data[idx]

dataset = dataset()
dataloader = torch.utils.data.DataLoader(dataset,
                                         batch_size=2,
                                         shuffle=False,
                                         num_workers=1)

for i, data_ in enumerate(dataloader):
    print(i)
    print(data_)

Run the observation ：

0
tensor([[1, 1, 1, 1],
        [2, 2, 2, 2]])
1
tensor([[3, 3, 3, 3],
        [4, 4, 4, 4]])
2
tensor([[5, 5, 5, 5],
        [6, 6, 6, 6]])

This example is enough to understand DataSet 了

2, dimension

Refer to this article ：
https://blog.csdn.net/xddwz/article/details/108405817

# -*- coding: utf-8 -*-
import numpy as np
import torch
from torch.utils.data.dataset import Dataset
from torch.utils.data.dataloader import DataLoader
from torchvision.transforms import transforms
import os
 
import cv2
from PIL import Image
 
 
class MyDataset(Dataset):
    def __init__(self, transform=None):
        self.transform = transforms.Compose([
            transforms.ToTensor()      #  Here is only the most basic example 
        ])
        self.image_path = './image_data2/'
        self.image_names = os.listdir(self.image_path)
 
    def __len__(self):
        return len(self.image_names)
 
    def __getitem__(self, item):
        image_name = self.image_names[item]
 
        image = cv2.imread(os.path.join(self.image_path, image_name))   #  What I read is BGR data 
        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)                  #  Turn into RGB, It can also be used. img = img[:, :, (2, 1, 0)]
        #  At this moment image yes H,W,C The order of , Therefore, the following needs to be transformed into C, H, W
        image = torch.from_numpy(image).permute(2, 0, 1)             
 
        # image = Image.open(os.path.join(self.image_path, image_name))
        # # print(image.shape)
        # image = self.transform(image)
        return image

This code is from the above link , The reason for moving here is to emphasize one thing ： We know torch The order of dimensions is BCHW, And in the above code __getitem__(), Is to return a picture , So what we need to pay attention to at this time is , The original dimensional order of a single picture is HWC, That is, the dimension is (height, width, channel), We need to adjust its dimension to (channel, height, width), And then back again .

meanwhile , According to the above code , Also need attention , Back to image Of shape No (1, channel, height, width), It is (channel, height, width),batch The corresponding dimension is __getitem__() There is no need to consider .

For some channel by 1 Pictures of the , Add if necessary channel dimension , So just squeeze(0) That's it .