`Imagenet2012CorruptedTest.test_download_and_prepare_as_dataset` is broken

Short description

Imagenet2012CorruptedTest.test_download_and_prepare_as_dataset test gives an error FileNotFoundError: [Errno 2] No such file or directory: 'imagemagick' despite ImageMagick being installed. After that particular issue is patched, the next error is AttributeError: module 'cv2' has no attribute 'CVX_8U'.

Environment information

• Operating Systems: both Ubuntu 22.04 (in docker with virtualenv) and NixOS 21.11 with nixpkgs on 'master'.

• Python version: 3.9.9

• tensorflow-datasets/tfds-nightly version: 4.4.0

• tensorflow/tf-nightly version: 2.7.0

• Does the issue still exists with the last tfds-nightly package (pip install --upgrade tfds-nightly) ? Yes.

Reproduction instructions

pytest ./tensorflow_datasets/image_classification/imagenet2012_corrupted_test.py

Expected behavior

The test succeeds.

Additional context

There seem to be two problems with the tensorflow_datasets/image_classification/corruptions.py implementation:

• Most (all?) distributions call ImageMagick binary convert (for v6.x) or magick (but typically with convert as a link to it, for v7.x), I'm not sure which one uses imagemagick binary name => imagemagick should be replaced with convert.
• The code uses the constant CVX_8U which, I believe, doesn't exist and is a typo? => CVX_8U should be replaced with CV_8U.

When both of these problems are fixed - the tests succeeds. See also the patch pkgs/development/python-modules/tensorflow-datasets/corruptions.patch in https://github.com/NixOS/nixpkgs/pull/154117 LMK if you want me to create a pull request against tensorflow-datasets with this patch applied.