yolo11で学習しているのでultralyticsのバージョンが古いとエラーになります
I'm learning with yolo11, so if the version of ultralytics is older, I will get an error.
recommend confidence threshold:<0.6
口の形状によって安定しないケースがあります
In some cases, it may not be stable depending on the shape of the mouth.
YOLO11nで学習したイラスト用の口検出モデルです:
データセットはtrainとval合わせて約1000枚です
マスク領域を口の中ではなく、唇を覆う様に指定しています
基本的に女性の画像しか学習していませんが、男性を含む画像ではどちらにもマスクしているので、女性限定というわけでも有りません
Mouth detection model for illustration trained on YOLO11n:
The dataset is about 1000 train and val images.
The mask area is specified to cover the lips, not the mouth.
Basically, only female images are studied, but images containing men are masked as well, so it is not limited to women.
Description
FAQ
Comments (6)
Thanks for adding this! Do you think it'll perform well on realistic faces?
The real style is not trained at all and will not work very well.
Furthermore, I don't have any real-style data sets, so it will take some time to add them, so don't expect much.
Correction:
Actual testing detected the real style as well.However.
since this is an unexpected case, please consider it as only possible to detect the real style as well.
I would love to see a realistic version trained, I was unable to get good results on my generations while using it. The existing lips/mouth bbox model is very hard to use I find, and the teeth seg model doesn't always detect, so a seg mouth model would be fantastic! Would you consider a bounty or buzz commission for the project at all?
Hello, is this a bbox or seg model? Not so sure where to put for ComfyUI, thanks!
Hello! This is a "seg" model.
Model Placement:
Place it in: ComfyUI\models\ultralytics\segm
Model Whitelist:
Add the model filename to:
ComfyUI\user\default\ComfyUI-Impact-Subpack\model-whitelist.txt
```
mouth_seg_11n.pt
```
Please note: I don't primarily use ComfyUI, so I cannot fully verify that this setup will work perfectly.