There is a set of Pixel Level transforms that is used in the work Benchmarking Neural Network Robustness to Common Corruptions and Perturbations

The authors also share the code => we can absorb some transforms that they have into the library.

https://github.com/hendrycks/robustness/blob/master/ImageNet-C/create_c/make_imagenet_c.py

• ShotNoise
• Defocus
• GlassBlur

Initial attempt failed, we need to fix the following issues:
SixLabors/ImageSharp#1302 (comment)

Disabling the two (likely) bugs with [ActiveIssue] might be also fine.

Description

The shape parameter in disk documentation (and maybe others using the same machinery) seems misleading to me. Based on the following phrasing, I would expect shape=None and the shape=dimensions_of_my_disk to lead to equal results. (but it doesn't)

Image shape which is used to determine the maximum extent