Analyse target image programmatically

[niels]

Hello all,

I’m looking for a way to programmatically determine if an image is a suitable image target.

On our platform users can upload any image, which we process with the image training package. Unfortunately, training always succeeds, even if the image is not trackable.

Is there any way to programatically infer trackability from the resulting .zpt file? Or is there maybe another tool/API that provides the image analysis functionality, like the one used in the ZapWorks Designer?

Thanks in advance,
Niels

[abiro]

Hello! You might find this thread to be of some insight. It explains how the analyzer in Designer works.

Image analyzer outside of Designer

Hi Atlas, The problem with the analyser is that it’s just scoring the image with a very basic metric (looking for a decent spread of “detail”) - the output of that is usually correlated with tracking quality, but not always. It can also encourage people to add additional detail where it’s not really needed (people desperately trying to get the whole image “green”). The problem there is people often then add some repetitive patterns to try to get a “green” score, but repetitive patterns are not…

I’m not sure of a way to programmatically determine the trackability of an image, but that’s something I’d also like to figure out.

(This is linked within the previous thread, but here’s another thread to look at: )

Allow transparent areas of trained image to be ignored when tracking

If we could use the transparent areas of a PNG as a way to tell Zap to ignore those regions/pixels, we could do some really funky stuff. It’d break us free of rectangular targets. (Yeah, you can kinda do round things like badges/buttons at the moment, but some fabrics behind the badge trip up the tracker because you can’t get it to ignore the corners) For example: we could use our logo as an image target - set against a transparent background - and then anywhere in the world the logo appears, …

[niels]

Thanks for the pointers @abiro

Even an approximation of tracking quality would already be a significant improvement over the current situation.