Vladimir Iglovikov

Something like this: ```python class RandomCropFromBorders(DualTransform): """Cuts out a rectangular portion of the image by randomly indenting each side. There is no resize of image at the end. Args: crop_value...

Can copy from https://github.com/layumi/Person_reID_baseline_pytorch/blob/84360b7b964dc4b9746dd89837333ed34c1208c5/random_erasing.py#L57

Good question. I think that it should affect only the images, but in this case, it will have a different spirit than the CourseDropout. It looks like it should be...

Feel free to contribute it. Something like PadToSquare

Pixelate in the paper == Dowscale with `cv2.INTER_NEAREST` in the library.

ShotNoise in paper == ISONoise in the Library

@bonlime what is the objection to this transformation? :)

Works for me.

What does this change do? We do not use `histogram_matching` in the rest of the file.

Could you please provide links to the description of Mosaic Augmentation?