Non-square image targets, training with transparent PNGs?


A question that’s cropped up a few times, and while I did a quick test that seemed to work, I wanted to get the official verdict.

If you need a non rectangular tracking image (eg a round badge/pin), how should we prepare the image for training? Obviously the issue is that with a round design, we don’t know what background the camera will be seeing in the “corners” of the tracked area.

I tried using a transparent PNG, and in my very brief tests it seemed to work: the tracking seemed to work OK regardless of what background the badge was sitting on. But I didn’t have time to fully test it.

Am I right - is that the correct approach? Does the training, and the subsequent recognition/tracking allow for parts of the image to be ignored?


Hi, i also used a PNG for a round tracking image, that was the top of a container… and tested a lot of time and it’s ok.


(wow oops this had been sitting open on my work PC over the weekend bc I forgot to post it)

Personally I haven’t tried a non-rectangular tracking image, but I have been working with an ever-changing tracking image (the design department at work is sometimes late to send me an updated file) and have discovered that as long as there’s enough visual information to latch onto, Zappar can track to images with some minor differences. It’s quite remarkable how much can change or be missing from a tracking image while still tracking correctly and accurately. I think that would translate well to the corners of your image being different due to tracking a circle. Obviously, the closer to identical the background is the tracking image (so in this case, probably a white or something with little/no visual information) the better, but if it works in testing, you should be ok