How does one specify device type and orientation? about eye-tracker

[hugochan]

@abhinavvadrevu What you suggested is an interesting point. However, I don't think that paper considered either device type or orientation and there is no such information in the dataset. I suggest you to talk to the authors of the paper for more information.

[abhinavvadrevu]

Ah, thanks @hugochan! Another quick question: The paper used 224x224 images for the face and eyes, but your implementation is 64x64. Did you compress the original dataset and find that accuracy was not affected?

[hugochan]

@abhinavvadrevu Due to the limitation of computation power, a much smaller dataset was used. You can download the original dataset here.

[abhinavvadrevu]

Ah, gotcha. I thought you meant fewer samples when you said "smaller dataset" in the README - not smaller images.

So have you found that the performance is significantly worse with smaller images? I ask because I'm considering retraining with the original dataset (I'll only do it if there's a significant performance boost in doing that!)

[hugochan]

@abhinavvadrevu Sorry, I don't have the comparison results. But it is worth a try.

[abhinavvadrevu]

No worries, I'll let you know what I find!