Welcome to computer vision / machine learning research!

It is absolutely not common in CV or ML to use such underhanded tricks. Peer review is allergic to it.

Is not the same scale cut out used in Scaled Yolov4 and previous research? (Though this one is on AP not mAP). It seems it is needed since the improvement would be hard to appreciate if it would show a 0-100 scale, since improvement in OR models are precisely not dramatic. https://openaccess.thecvf.com/content/CVPR2021/papers/Wang_S...

I honestly find that deceiving and I would’ve said so in a review. Since it /is/ a percentage, I think you should absolutely show the zero - and the maximum if applicable. The correct scale is 0-1. It being a minute difference is exactly the type of thing a fair plot would reveal.

However, I think it’s better to use a semilog plot here - and show 1 - AP. I’m sure there’s a name for that number too. Then perhaps invert the Y axis if you really want the curve trending upward. No need for the cut, and infinite precision.

You of course should note this in the figure caption. This goes doubly so for cutting axes.

How about cherry picking (which peer review can't easily catch)?

IMO being able to cherry-pick results is pretty interesting in itself. Also, good works tend to come with online demos or even exact parameters — hard to hide your shortcomings when people can reproduce your work.