Hi,
Was wondering about which cases to remove when compiling a faceset, particularly the pornstar faceset. I have more than enough (5k+) pictures of straight-on faces, faces at a slight angle all while talking etc. I have enough of those.
But what about cases like these:
[img=128x128] Only 1 eye visible
[img=128x128] Blowjob obstruction
[img=128x128] Side of face obstructed / deformed
[img=128x128] Slight hair obstruction, chin not visible
[img=128x128] Hair obstruction from the side/edge of face
[img=128x128]https://i.imgur.com/ABCNwtS.png[/img] Forehead not visible
So the first instinct is to just remove all of these. But should the training data really just be straights-on shots? Like at what point do I decide the angle is too extreme to include?
Was wondering about which cases to remove when compiling a faceset, particularly the pornstar faceset. I have more than enough (5k+) pictures of straight-on faces, faces at a slight angle all while talking etc. I have enough of those.
But what about cases like these:
[img=128x128] Only 1 eye visible
[img=128x128] Blowjob obstruction
[img=128x128] Side of face obstructed / deformed
[img=128x128] Slight hair obstruction, chin not visible
[img=128x128] Hair obstruction from the side/edge of face
[img=128x128]https://i.imgur.com/ABCNwtS.png[/img] Forehead not visible
So the first instinct is to just remove all of these. But should the training data really just be straights-on shots? Like at what point do I decide the angle is too extreme to include?