I'm working on some applications that require matching real world items to augmentations. I'm doing this by defining offsets from an image target. My issue is that the recognized pose of the target doesn't seem to be very consistent. Sometimes the augmentations relative to the target look great and another time, without changing anything, they're off by a couple inches. Looking at the target longer seems to help but it's not clear. I've seen this behaviour on iOS, Android, and HoloLens.
I've tried some different targets but the current one is 5"x3" and has a 5 star rating. It's not printed on glossy paper. The lighting is good and also out of my control.
I'm wondering if anyone has any tips for improving the consistency of the image target pose. Should I make the users keep the marker in view for 5 seconds? Any other techiniques?