issue of eye closed img

[lumos1iu]

When detecting a closed-eye portrait, the model still seems to inference according to the open-eye relationship. The landmarks of eyes is too big, Two examples are like this:

May I ask, is the model overconfident about the eye-open keypoint interrelationships. If the number of closed-eye portraits in the dataset is increased, in principle do you think this problem can be solved?


thank u very much！

[Jiahao-UTS]

Maybe your dataset contains too few faces with closed eyes.