Closed daladulu closed 1 year ago
The complete failure is likely due to am improper face bounding box. I suggest visualizing them too, and if necessary switch to a different face detector (3 options are available directly within the library).
Regarding the eyes, this is unfortunately a consequence of the training data used, which had very few images where the eyes were closed.
Sorry, there are some problems. Incorrect landmarks, especially for the images with eyes closed.
![plot1_217](https://user-images.githubusercontent.com/50403079/214844754-9f18ca96-8e8c-44c0-8f92-19dd53ccd0e7.png)