take EXIF orientation tags into account when fixing corrupt images (#5270)

* take EXIF orientation tags into account when fixing corrupt images

* fit 120 char

* sort imports

* Update local exif_transpose comment

We have a local inplace version that is faster than the official as the image is not copied. AutoShape() uses this for Hub models, but here it is not important as the datasets.py usage is infrequent (AutoShape() it is applied every image).

* Update datasets.py

Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

utils/datasets.py

@@ -22,7 +22,7 @@ import numpy as np
 import torch
 import torch.nn.functional as F
 import yaml
-from PIL import Image, ExifTags
+from PIL import Image, ImageOps, ExifTags
 from torch.utils.data import Dataset
 from tqdm import tqdm
 
@@ -69,7 +69,7 @@ def exif_size(img):
 def exif_transpose(image):
     """
     Transpose a PIL image accordingly if it has an EXIF Orientation tag.
-    From https://github.com/python-pillow/Pillow/blob/master/src/PIL/ImageOps.py
+    Inplace version of https://github.com/python-pillow/Pillow/blob/master/src/PIL/ImageOps.py exif_transpose()
 
     :param image: The image to transpose.
     :return: An image.
@@ -896,7 +896,7 @@ def verify_image_label(args):
             with open(im_file, 'rb') as f:
                 f.seek(-2, 2)
                 if f.read() != b'\xff\xd9':  # corrupt JPEG
-                    Image.open(im_file).save(im_file, format='JPEG', subsampling=0, quality=100)  # re-save image
+                    ImageOps.exif_transpose(Image.open(im_file)).save(im_file, 'JPEG', subsampling=0, quality=100)
                     msg = f'{prefix}WARNING: {im_file}: corrupt JPEG restored and saved'
 
         # verify labels