I am doing some research on gaze estimation model L2CS provided in here. When you look at the inference predictions on provided image, when face/eye direction barely changes, inference predictions vary a lot sometimes. I was wondering if I can improve it a bit. I was wondering about kalman filter but I will lose some fps on it. Are there some changes on architectural level I can do ? Or maybe something else ?

Does any one faced similar issue ? Thanks !



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.