We've been working with the Vuforia Hololens Sample 1 & 2 asset available on the Unity Asset Store for some time now, and while we have been able to get image tracking working in a basic form, we've discovered a strange offset issue that we think is being caused during the initialisation of some part of the Vuforia plugin.
Intermittently, we have encountered what appears to be a random rotational offset in the position of the virtual image target in relation to ise physical print counterpart. Through some experimentation, we discovered that it in fact seems to relate to how much the user changed the orientation & position of their head while the app is loading.
To test this, we set up a basic project and ran a few tests where we would start the app, wait for the 2D unity window to display the Vuforia splash screen, and then turn some significant rotation amount to the left or right from our starting position. Once the app finishes loading we would look at the image target and observe an offset that seemed representative of how much we turned during that app's load.
I've attached a zip to this post that should help to diagnose this issue. It contains a Unity project demonstrating our setup - duplicated directly from the Image Targets sample scene in the framework, along with an AppX demonstrating the issue that is ready for side-loading.
The best way we've found to replicate this bug is to swing your head from side to side as the app loads, to maximise the chance that there will be a random-ish offset once the app has finished loading and the app is ready to track image targets.
Please note that we have seen instances of this issuein HoloLens since as far back as early 2018, when we first started using Vuforia in Hololens apps, so we're aware that is is a long-standing issue. However, we have also used Vuforia in other platforms such as iOS & Android, and we have not encountered this issue on those platforms, so we believe it to be specific to the Vuforia framework(s) for Hololens.
As our client will be deploying these devices to trainees who are not expected to be familiar with the idiosyncrasies of this technology, we need to be able to offer them a solution more robust than the workaround of them telling their trainees to stand still as a statue while the app loads in their headset. So any assistance you can provide to help us fix or bypass this bug would be appreciated!