將 python 列表轉換為張量 pytorch

Question

我想將像素值列表轉換為張量，但出現錯誤。 我的代碼計算圖像中每個檢測到的 object 的像素值 (RGB)。 我們如何將列表轉換為張量？

我的代碼：

         cropped_images =[]

         imgs = PIL.Image.open(img_path).convert('RGB')
         #print(img_path)
         image_width, image_height = imgs.size
         imgArrays = np.array(imgs)

                 
         X = (xCenter*image_width)         
         Y = (yCenter*image_height)         
         W = (Width*image_width)          
         H = (Height*image_height)         
         cropped_image = np.zeros((image_height, image_width))
                        
         for i in range(len(X)):
                 x1, y1, w, h = X[i], Y[i], W[i], H[i]
                 x_start = int(x1 - (w/2))
                 y_start = int(y1 - (h/2))
                 x_end = int(x_start + w)
                 y_end = int(y_start + h)
                              
                 temp = imgArrays[y_start: y_end, x_start: x_end]

                 cropped_image_pixels = torch.as_tensor(temp) 
                 cropped_images.append(cropped_image_pixels)
                 stacked_tensor = torch.stack(cropped_images)

                 print(stacked_tensor)

錯誤：

RuntimeError                              Traceback (most recent call last)
<ipython-input-82-653a155c3b71> in <module>()
    130 
    131 if __name__=="__main__":
--> 132     main()

2 frames
<ipython-input-80-670335a0656c> in __getitem__(self, idx)
     76                  cropped_image_pixels = torch.as_tensor(temp)
     77                  cropped_images.append(cropped_image_pixels)
---> 78                  stacked_tensor = torch.stack(cropped_images)
     79 
     80                  print(stacked_tensor)

RuntimeError: stack expects each tensor to be equal size, but got [506, 343, 3] at entry 0 and [520, 334, 3] at entry 1

Answer 1

張量列表有兩個張量，很明顯兩者的大小不同

torch.stack(tensors, dim=0, *, out=None) → Tensor
Concatenates a sequence of tensors along a new dimension.

All tensors need to be of the same size.

你可以使用這個偽代碼

import torchvision.transforms as transforms
.
.
.
.
temp=[]
for img_name in LIST:
    img=cv2.resize(img,(H,W))
    temp.append(img)

train_x=np.asarray(temp)

transform = transforms.Compose(
    [transforms.ToTensor(),

檢查文檔