问题再现

from torchvision.io.image import read_image
img_path = 'border-collie.jpg'
img = read_image(img_path)

报错如下：

WARNING:root:no value was provided for `target_layer`, thus set to 'layer4'.
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
File ~/anaconda3/envs/DVS/lib/python3.8/site-packages/torch/_ops.py:757, in _OpNamespace.__getattr__(self, op_name)756 try:
--> 757     op, overload_names = torch._C._jit_get_operation(qualified_op_name)758 except RuntimeError as e:759     # Turn this into AttributeError so getattr(obj, key, default)760     # works (this is called by TorchScript with __origin__)RuntimeError: No such operator image::read_fileThe above exception was the direct cause of the following exception:AttributeError                            Traceback (most recent call last)
/home/algroup/chenao/RankMatch/CAM.ipynb 单元格 16 in ()1 # Get your input
----> 2 img = read_image(img_path)3 # Preprocess it for your chosen model4 input_tensor = normalize(resize(img, (224, 224)) / 255., [0.485, 0.456, 0.406], [0.229, 0.224, 0.225])File ~/anaconda3/envs/DVS/lib/python3.8/site-packages/torchvision/io/image.py:222, in read_image(path, mode)206 def read_image(path: str, mode: ImageReadMode = ImageReadMode.UNCHANGED) -> torch.Tensor:207     """208     Reads a JPEG or PNG image into a 3 dimensional RGB Tensor.209     Optionally converts the image to the desired format.(...)220         output (Tensor[image_channels, image_height, image_width])221     """
--> 222     data = read_file(path)223     return decode_image(data, mode)File ~/anaconda3/envs/DVS/lib/python3.8/site-packages/torchvision/io/image.py:42, in read_file(path)31 def read_file(path: str) -> torch.Tensor:32     """33     Reads and outputs the bytes contents of a file as a uint8 Tensor34     with one dimension.(...)40         data (Tensor)41     """
---> 42     data = torch.ops.image.read_file(path)43     return dataFile ~/anaconda3/envs/DVS/lib/python3.8/site-packages/torch/_ops.py:761, in _OpNamespace.__getattr__(self, op_name)757     op, overload_names = torch._C._jit_get_operation(qualified_op_name)758 except RuntimeError as e:759     # Turn this into AttributeError so getattr(obj, key, default)760     # works (this is called by TorchScript with __origin__)
--> 761     raise AttributeError(762         f"'_OpNamespace' '{self.name}' object has no attribute '{op_name}'"763     ) from e765 # let the script frontend know that op is identical to the builtin op766 # with qualified_op_name767 torch.jit._builtins._register_builtin(op, qualified_op_name)AttributeError: '_OpNamespace' 'image' object has no attribute 'read_file'
# packages in environment at /home/algroup/anaconda3/envs/DVS:
#
# Name                    Version                   Build  Channel
torch                     2.1.1                    pypi_0    pypi
torchcam                  0.4.0                    pypi_0    pypi
torchextractor            0.3.0                    pypi_0    pypi
torchvision               0.11.3                   pypi_0    pypiNote: you may need to restart the kernel to use updated packages.

修改版本

说是torchvision版本不匹配什么的，我懒得改，因为还和pytorch有关系，要改要一起改很麻烦；

改用Image.open()读取文件

from PIL import Image
import torchvision.transforms as transformstotensor = transforms.ToTensor()
img = totensor(Image.open(path))

对于菜鸡而言，不敢瞎改，怕改错了整个程序都崩溃了。但是读清楚之后，发现出问题的只是一个读取图片的小函数，之前理论上如果能运行的话，是吧.jpg文件读取成tensor，那我用别的函数做同样的时候就好，这是就自然想到Image.open()把图片转换成什么类型了，如果不是tensor，还要加一道转换步骤，就这样逻辑就严密了！

参考

[1] No such operator image::read_file问题解决