mmcv, geometry.py not compatible with latest Pillow 9.1.0 (and possibly earlier versions)

Hello,

I am using MMDeploy to generate an optimized model trained by MMDetection.

It looks like there is an incompatibility with latest Pillow. I get the following error when executing MMDeploy tools/deploy.py. The error occurs in mmcv geometric.py

Traceback (most recent call last):
  File "tools/deploy.py", line 7, in <module>
    import mmcv
  File "/home/nvidia/git/project/venv-mmdet/lib/python3.6/site-packages/mmcv/__init__.py", line 5, in <module>
    from .image import *
  File "/home/nvidia/git/project/venv-mmdet/lib/python3.6/site-packages/mmcv/image/__init__.py", line 5, in <module>
    from .geometric import (cutout, imcrop, imflip, imflip_, impad,
  File "/home/nvidia/git/project/venv-mmdet/lib/python3.6/site-packages/mmcv/image/geometric.py", line 42, in <module>
    'nearest': Image.NEAREST,
AttributeError: module 'PIL.Image' has no attribute 'NEAREST'

This error does not occur on an earlier version of Pillow (7.0.0) but it occurs on 9.1.0. I can replicate the issue by calling Python3 directly:

• Pillow 7.0.0

>>> from PIL import Image
>>> Image.NEAREST
0

• Pillow 9.1.0

from PIL import Image
>>> Image.NEAREST
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'PIL.Image' has no attribute 'NEAREST'

Solution: It looks like the NEAREST… variables has been moved into PIL.Image.Resampling: PIL.Image.Resampling.NEAREST

Alternatively, the deprecated filters can be accessed using the following names:

PIL.Image.NONE = Resampling.NEAREST
PIL.Image.LINEAR = Resampling.BILINEAR
PIL.Image.CUBIC = Resampling.BICUBIC
PIL.Image.ANTIALIAS = Resampling.LANCZOS

[tehkillerbee]

@imabackstabber Sure, I can try to do a PR. I assume we still need to be backwards compatible with the older versions of pillow?

[imabackstabber]

@tehkillerbee Thanks for prompted reply! And Also, feel free to let us know if there’s any problem when updating and we’re always ready for help contributing!