Towards mitigating uncann(eye)ness in face swaps via gaze-centric loss terms