Coordinates in sanakoyeu_et_al_2018.RandomCrop.generate_anchors() are interchanged

[pmeier]

import torch
import pystiche_papers.sanakoyeu_et_al_2018 as paper

crop_width = 100
crop_height = 1

torch.manual_seed(0)
image_width = crop_width * 2
image_height = crop_height * 2

anchors = paper.RandomCrop.generate_anchors(
    batch_size=1,
    image_size=(image_height, image_width),
    crop_size=(crop_height, crop_width),
    same_on_batch=True,
)
anchor = anchors[0, 0, :]
print(anchor)

x, y = anchor.tolist()
assert x < image_width, f"{x} >= {image_width}"
assert y < image_height, f"{y} >= {image_height}"

tensor([ 0, 77], dtype=torch.int32)
AssertionError: 77 >= 2

This happens because the image and crop size place the height before the width (as is the default in PyTorch)

https://github.com/pmeier/pystiche_papers/blob/f47871f9334d21adc27de979c2480274bb73170e/pystiche_papers/sanakoyeu_et_al_2018/_data.py#L194-L200

whereas kornia.augmentation.functional.apply_crop expects the horizontal coordinate before the vertical.

Cc @jbueltemeier