The Eye Tribe

Hi, I just received the device, happy to test it !

First thing: the device itself is really great: small (the smallest I've tested until now), well crafted, nice finish. Great!

The USB cable is a bit too heavy for the tripod, if the cable lay on the table the device will fall on its side. But that's the only default I've noticed so far.

The tutorial is great, calibration runs smoothly and the tracking seems to work pretty well. I tested it with glasses on a surface pro 2 and I still have a fairly good tracking. The mouse support is smooth, but the cursor don't stay fixed on the gaze point (it continue to move until it touch the side of the screen). Not a big issue for a pre release, it's already pretty good, congrats!

I just downloaded the SDK, I'll integrate your client in our NUI test project to give you more feedback on the development kit.

Promising eye tracking solution, for this price range this is really amazing, thanks!

antoine

Product wise:

[*] I found my glasses did not cooperate with the eye tribe too well unfortunately.
[*] The USB cable does tend to tilt the sensor unless supported. I would perhaps recommend a cable that comes out from the sensor bar that is a female for the usb port that the heavy cable can then be plugged into. I can see why it isn't located on the back or the bottom for laptop support.
[*] The packaging is great!
[*] The software looks great! Eye tribe server should minimize to the small icon notification tray for finished product to free up task bar.
[*] Consider having an icon for the app that changes color based on the accuracy (like the colour behind the eyes on the eye tribe UI app does)

I think you would have a one up on competition if you also looked into bluetooth based device that would be like a ring or something to put on the pointer finger for clicking which would free up the hand to just work on keyboard and not block any keyboard keys.

I do believe you have an amazing product, especially for the price range! I was initially looking at the Tobii and their price even with the CES discount didn't meet up with yours!

Developer wise:
[*] I would like to see individual normalized eye center positions in addition to screen pixels. I think this may be coming from pupil center coordinates but not sure yet. The reason I ask is because I want to add support for multiple monitors.
[*] I think you need an additional calibration step to adjust the vertical misalignment based on pupil size or distance between eyes when the user either moves closer or farther away. I recommend having them look at a single point and sit back, and then looking at the same point have them lean forward.
[*] It would be interesting to see head tracking position included with the GazeData class based on the observed distance between the pupils (I can do this myself, but consider it as an extension method to return the calculated z position perhaps?)

Hi Antoine & Kevin,

Thanks for the great feedback. I will certainly take it into consideration for our development roadmap.

We spend a lot of time working on the device itself and I'm happy to hear that you notice. It's certainly a quality piece of hardware that's selling way below it's value. The strength and precision of the CNC milling is very high and the post processing surface work (multiple anodizing treatments) gives a really strong and scratch resistant surface. We've done many drop-tests, it wouldn't break until we dropped it from 4 meters into a stone floor, even so the only damage done was that the lens mount came loose (all electronics still working).

I know that the USB 3.0 cable is pretty heavy, we've tried several options here, some were thinner and lighter but felt flimsy and I believe we would have seen signaling issues using them. One option would have been to ship with a typical round cable but we found that it created an even stronger pull on the device. This could been solved by making the device heavier but then we're heading down the wrong path. My immediate thought would be to route the USB cable around the monitor or along the table to give it some support. Long term I believe the right solution is to have a better mounting mechanism and that's something we are working on at the moment, small brackets/mounts tailored for each form factor (laptop, tablet and desktop monitor).

Kevin,
Good idea regarding the notification tray, I've seen this implemented in the Leap Motion software and it works pretty well. I also like your idea about the color indication for the icon so that you immediately could see the tracking quality without having to switch to the app.

As for the head/distance feedback I hope that I've addressed some of it in this post.

By placing the tripod with two legs back and one in the front strengthens it dramatically. Se image from Martin here: viewtopic.php?f=9&t=19

I just received the EyeTribe !!!

My comments:

a) Firstly, it is awesome..

b) I tested it with laptop and pc.
The calibration result is excellent with laptop
The calibration result is poor or moderate with pc. (somewhere i do a mistake)

c) I used the option "Mouse gaze redirect" to test the correlation between my gaze and mouse position. I opened a pdf file and read the text (font : Arial 12 and zoom out 120%).
The result at X axis is fantastic!! I didn't remain such good result!!
But the result at Y axis is not accurate. Has someone an answer about this behavior?

d) If there was a support with java samples would be very helpful..

Finally, I am very satisfied with the device. .

Regards,
Giannis

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...