why skimage.io.imread warns “corrupt exif data” but cv2.imread doesn't?

Question

I am using pytorch to build a neural network classifier.

I read somewhere that many python packages use the same c++ api when reading a picture, so I thought skimage.io.imread and cv2.imread would be the same except channel order. However I got warning like this in skimage,

/xxx/lib/python3.6/site-packages/PIL/TiffImagePlugin.py:766: UserWarning: Possibly corrupt EXIF data.  Expecting to read 1245184 bytes but only got 0. Skipping tag 0
  " Skipping tag %s" % (size, len(data), tag)
/xxx/lib/python3.6/site-packages/PIL/TiffImagePlugin.py:785: UserWarning: Corrupt EXIF data.  Expecting to read 12 bytes but only got 11. 
  warnings.warn(str(msg))

but cv2 works without warnings.

My code goes like this:

• in Dataset

    ...

    def __getitem__(self, index):
        file = os.path.join(self.root_dir, self.files[index])
        image = io.imread(file)
        # image = cv2.imread(file)
        if self.transform:
            image = self.transform(image)
        return image

• transforms:

transform = transforms.Compose([
    transforms.ToPILImage(),
    transforms.Resize((128, 128)),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor()
])