如何在pytorch对象检测中添加转换

Question

我是 PyTorch 的新手，正在浏览 PyTorch 对象检测文档教程pytorch docx 。 在他们的合作版本中，我进行了以下更改以添加一些转换技术。

1. 首先修改 PennFudanDataset(torch.utils.data.Dataset) 类的__getitem__方法

if self.transforms is not None:
   img = self.transforms(img)     
   target = T.ToTensor()(target)
   return img, target

In actual documentation it is 
if self.transforms is not None:
   img, target = self.transforms(img, target)  

其次，在get_transform(train)函数处。

def get_transform(train):
  if train:
    transformed = T.Compose([             
           T.ToTensor(),
           T.GaussianBlur(kernel_size=5, sigma=(0.1, 2.0)),
          T.ColorJitter(brightness=[0.1, 0.2], contrast=[0.1, 0.2], saturation=[0, 0.2], hue=[0,0.5])
    ])
    return transformed

  else:
    return T.ToTensor()

**In the documentation it is-** 
def get_transform(train):
    transforms = []
    transforms.append(T.ToTensor())
    if train:
        transforms.append(T.RandomHorizontalFlip(0.5))
    return T.Compose(transforms)

在实现代码时，我收到以下错误。 我无法得到我做错了什么。

TypeError: Caught TypeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/worker.py", line 198, in _worker_loop
    data = fetcher.fetch(index)
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/fetch.py", line 44, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataset.py", line 272, in __getitem__
    return self.dataset[self.indices[idx]]
  File "<ipython-input-41-94e93ff7a132>", line 72, in __getitem__
    target = T.ToTensor()(target)
  File "/usr/local/lib/python3.6/dist-packages/torchvision/transforms/transforms.py", line 104, in __call__
    return F.to_tensor(pic)
  File "/usr/local/lib/python3.6/dist-packages/torchvision/transforms/functional.py", line 64, in to_tensor
    raise TypeError('pic should be PIL Image or ndarray. Got {}'.format(type(pic)))
TypeError: pic should be PIL Image or ndarray. Got <class 'dict'>

Answer 1

我相信 Pytorch 转换仅适用于图像（在这种情况下为 PIL 图像或 np 数组），而不适用于标签（根据跟踪是字典）。 因此，我认为您不需要像在__getitem__函数中的这一行target = T.ToTensor()(target)那样“张量化”标签。