You may have noticed that the landmark estimates are pretty noisy. That may hamper further processing.
Of course we can add some low-pass filtering on the frames directly or on the landmark data.
Should filtering be implemented in e.g. Irish_Mesh and Face_Mesh, or do we want to filter in some higher level functional block, such as gaze orientation estimation or head-pose estimation? Or perhaps both? E.g. some image smoothing in the lower nodes and data smoothing in higher nodes.
I would like to learn you preference.
You may have noticed that the landmark estimates are pretty noisy. That may hamper further processing.
Of course we can add some low-pass filtering on the frames directly or on the landmark data.
Should filtering be implemented in e.g. Irish_Mesh and Face_Mesh, or do we want to filter in some higher level functional block, such as gaze orientation estimation or head-pose estimation? Or perhaps both? E.g. some image smoothing in the lower nodes and data smoothing in higher nodes.
I would like to learn you preference.