I just gave it a try. I first used Go Eye's integrated browser to find a nice image from Google (it is probably a computer generated one but it looks real enough).
Long tap the image to activate the 'Recognize' button. After tapping it I got a warning that Go Eye could not recognize it automatically. Go Eye can automatically recognize two-dimensional computer generated Go board images (e.g. a screen capture of KGS client, or a go problem one can find a Sensei's library). Because now the input is a 3D image, Go eye needs some extra help.
As shown in the second screen capture, there are 4 green circles to help identify the 4 corners of the board grid. I found this fairly easy to use even on my iPhone 5's tiny screen. Once I tap the green circle I got a magnified view so that the 4 1-1 points can be accurately marked.
The last screen capture shows the recognized board position. Stones at B18, M9 and a few at the bottom right corners were missing. Also the color of K1 and Q2 were wrong. The majority of stones were recognized correctly.
Overall I think for a first release of real board recognition, it works reasonably well. Since image recognition is a very challenging task and can never guarantee to be 100% successful, it would be useful to provide a more convenient way to allow easy editing of the recognized image (now I can click the X button to edit but iPhone screen is just too small so it is a painful practice).
I also think the way using magnifier to locate board corners is very intuitive. Maybe it could be used as an alternative means to put stones on board.
Anyway, it is so happy to get this (though there is one faulty stone at bottom).