Pytorch从一个输入目录中加载所有的PNG图像，并将它们存储在张量中

 1 import os
 2 import imageio
 3 from imageio import imread
 4 import torch
 5 
 6 # batch_size = 3
 7 # batch = torch.zeros(batch_size, 3, 256, 256, dtype=torch.uint8)
 8 # batch.shape  #torch.Size([3, 3, 256, 256])[B,C,H,W]
 9 
10 data_dir = 'C:/Users/Dell/Pictures/'
11 filenames = [name for name in os.listdir(data_dir)
12             if os.path.splitext(name)[-1] == '.png'] #选择指定目录下的.png图片
13 
14 for i, filename in enumerate(filenames):
15     img_arr = imageio.imread(os.path.join(data_dir, filename)) #imread读入为H*W*C
16     img_t = torch.from_numpy(img_arr)
17     img_t = img_t.permute(2, 0, 1) #交换维度
18     img_t = img_t[:3]  #只保留前3个通道
19     batch[i] = img_t   #指第i个维度上的所有数据