Technically not a new release, but i haven’t officially announced it before.
I know quite a few people use my yolo models, so i thought it’s a good time to let them know there is an update 😀
I have published new version of my Face Segmentation model some time ago, you can find it here – https://huggingface.co/Anzhc/Anzhcs_YOLOs#face-segmentation – you also can read about it more there.
Alternatively, direct download link – https://huggingface.co/Anzhc/Anzhcs_YOLOs/blob/main/Anzhc%20Face%20seg%20640%20v3%20y11n.pt
What changed?
– Reworked dataset.
Old dataset was aiming at accurate segmentation while avoiding hair, which left some people unsatisfied, because eyebrows are often covered, so emotion inpaint could be more complicated.
New dataset targets area with eyebrows included, which should improve your adetailing experience.
– Better performance.
Particularly in more challenging situations, usually new version detects more faces and better.
What this can be used for?
Primarily it is being made as a model for Adetailer, to replace default YOLO face detection, which provides only bbox. Segmentation model provides a polygon, which creates much more accurate mask, that allows for much less obvious seams, if any.
Other than that, depends on your workflow.
Currently dataset is actually quite compact, so there is a large room for improvement.
Absolutely coincidentally, im also about to stream some data annotation for that model, to prepare v4.
I will answer comments after stream, but if you want me to answer your questions in real time, or just wanna see how data for YOLOs is being made, i welcome you here – https://www.twitch.tv/anzhc
(p.s. there is nothing actually interesting happening, it really is only if you want to ask stuff)
