Shape detector was returning them that way this whole time, but localizer assumes the xy is the center of the target. It was probably like this because the testing data has its labels in top-left xy plus width/height format. I could've also fixed this by converting the top left to center coordinates every time I use the localizer, but I feel like this is cleaner.
Summary
Shape detector was returning them that way this whole time, but localizer assumes the xy is the center of the target. It was probably like this because the testing data has its labels in top-left xy plus width/height format. I could've also fixed this by converting the top left to center coordinates every time I use the localizer, but I feel like this is cleaner.