My first go with the device was also very smooth. Not a single error or issue. (Windows 7 64-bit). The biggest hurdle was to squeeze myself under my desk (and its steel support structures) in order to get to the USB3 port on the back of my PC
The first go at calibration gave 4, even on my 30" main monitor (bigger than the recommended 24").
For the second calibration, after being able to see the "eyes" on the UI, I noticed the camera had to be slightly moved to left in order to get the eyes centered. I guess the camera is not in the center of the device. (Could be nice to have e.g. some mark on the back of the device showing the camera spot, sort of "put this spot against the logo on the center of your monitor bottom frame (if there is one)".) Further calibrations only gave 3, but I was possibly also a little bit further away. However, the tracking became a little bit more accurate anyway; after the first calibration, the left corner circles were flickering a bit, now they do not.
Third calibration (after a while of "playing with it"), and adjusting the "area size" to 2000x1600, instead of the normal 2560x1600, gave 4 stars, and very well working calibration. I guess I have to play with that a bit and see what will give best results.
The biggest semi-issue with the calibration process was that it may be a bit too eager; it will start shrinking the circles as soon as the eyes are even about there. But for me, it takes about 0.5-1.0 sec before my eyes are well focused and aiming at the spot. I would add a new parameter to the calibration UI: "point sample delay"; when the calibration routine notices the eyes are at the circle, it would wait that amount of time (of continuous "hit") before starting collecting the samples or shrinking the circle. (Well, it could already do such delay, but if so, let that delay show during the calibration, too. Or perhaps it just throws away the values that are most distant from the average or something.) Optionally, let the user press key or mouse button when he/she thinks "now I'm gazing that spot, start sampling".
I also used the mouse gaze redirect to check how it works in real time. I, too, noticed the difference between vertical and horizontal accuracy. But it was more like so that with horizontal shift or turn of head, the cursor first jumped correspondingly, but it quite quickly wandered closer to the right spot. For vertical turn or shift, the error just kept there. This could have something to do with having two eyes horizontally, which should allow some ability to calculate away other horizontal variables; no such luck with vertical direction.
The vertical changes can be reduced by instructing the user to sit with the back well against the chair, that should sort of "lock" the back, neck, and head to roughly static height. At least if the back of the chair doesn't yield. Static position is not good for long term ergonomics, though, so better have frequent breaks of stretching.
Remark for the installation: I know this is just an early SDK / dev-stuff, but for future, make sure the installer does not ask for complete control / full permissions. It does not (or at least should not) really need that much. Makes any security aware person go ballistic... For example, I wouldn't probably get a permission to install this thing at my work due to that. We could use it e.g. for website and user interface analysis.
And the license for this dev-grade software... Did anyone read it? Mostly quite normal, but there are two spots that made me think a moment. Strictly taken, I think I am breaking that license at the moment, as, while I am a developer, I am personally not a commercial one
That tiny detail wasn't revealed by the website "terms of sale". Though, I'd think The EyeTribe would welcome any non-commercial developers (and thus applications), too... Another issue with the license was the part that mentioned something about using on a single something (I forgot the exact words used). Would that still allow
installing the software on multiple machines, as long as only one is being used at a time, unless having multiple devices? The devil is in the details. At least if you ask any lawyer
.
All in all, seems good so far, though the amount cursor jitter could indicate that my main early idea may be at a risk. The second idea should still be good to go. Shall see...
