Scenario: I take the scanner paper I printed, place a CFL bulb on it and go through the full scanning process. This works fine, but when I test, the test only works while the CFL is on the scanning paper. If I take it off it does not recognize. The same happens after I deploy the app to my device. From what I understand the scanner should not consider the paper part of the object features. The normal image recognition works great, even more so if I slim down on the unimportant features in an image editor before I upload, but using full 3d objects just won't work properly.
I have one more issue past this in the same realm: While recognizing an object, it only works well from the initial face (I usually start top down). If I rotate the object and keep it on the paper while running the test it has serious troubles recognizing the object, but the main face takes only a second or two.
I have tried using this with three android devices including a galaxy phone.