Box format degeneracy has been a footgun for computer vision developers since forever. You can define a rectangle as two corner coordinates, a coordinate + width + height. Since "one coordinate" can be a corner or the center, there are normally 6 variations and every single one exists somewhere. This also causes havoc for validation because it's easy to forget and wonder why your metrics are all practically zero because you didn't specify the right one.

There's a neat table here: https://dragoneye.ai/blog/a-guide-to-bounding-box-formats

Picking yxyx was certainly a decision.