Open EderSant opened 5 years ago
@antoinelame After I called the is_top and is_low functions in example.py it no longer detects the left or right looks. Most of the time it's in the center or the bottom.
What if instead of positions there was a way to delimit a quadrant of the screen for each direction? In this way, the margin of error will not be reduced?
Hi @EderSant
First of all, what accuracy do you get on the pupils detection? I'm not talking about the gaze direction (left, right, top, bottom) but just about the coordinates you get and that locate your pupils. A good accuracy on that is the first requirement to get a good prediction of the gaze direction.
If you have a wrong accuracy on the pupil detection, we can work on that. There is already an automatic calibration system, but it's sure that it can still be improved.
I see that you wrote is_top()
and is_low()
functions using vertical_ratio()
, well done. It can't have any incidence on the right
and left
detection, or maybe you wrote an implementation with wrong conditions. Please, share the code you wrote using the library.
The library is here to provide two things:
Then, it's up to library users to build with it their own implementation, like a quadrant in your case.
Hello Antoinelame,
Sorry for the delay in replying, I was working on another part of the code. On the coordinates, what I usually have to use is:
25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25] [20, 20, 20, 20, 20, 20, 20, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25]
Hello @antoinelame
What do you suggest to solve the problem with the margin of error, especially with the top and bottom positions? I implemented them in example.py and removed the blinking function, however the margin of error is very large.