The position of the detected face depends on the part of the face that receives the light.
Proportions should ideally depend on features, or history, rather than the detected Haar/LBP position, especially if no eyes were detected in the expected ROI.
Should we have a backup ROI dependent on the previous location of the eyes, or just use the whole face in that case?
With the new hierarchical approach on the ROI, this is not as likely to happen. This item will be closed, and might be reopened of the bug is reproduced.
The position of the detected face depends on the part of the face that receives the light.
Proportions should ideally depend on features, or history, rather than the detected Haar/LBP position, especially if no eyes were detected in the expected ROI.
Should we have a backup ROI dependent on the previous location of the eyes, or just use the whole face in that case?