Using adetailerfootyolov8x.pt as a starting point made by https://civarchive.com/user/Monet_Einsley, I trained this model. It uses ReignSolesLora training dataset, so the model is highly focused on detecting the soles of the feet. It works incredibly well with soles generated from said Lora, but can also detect the soles of a woman's feet in other anime style images.
This is a BBOX type model made to work with Adetailer, ComfyUI (needs a compatible workflow), or some other Ultralytics program. Unfortunately I could not find a way to convert it to .safetensors, so the .pt file is the best I can do.
I recommend using it with a confidence threshold of .7 or lower if you find it isn't recognizing the soles in your image.
Description
First version, trained on about 300 total anime style images (200 for training, 100 for validation). Used the basic BBOX training guide https://civitai.com/articles/1224/training-a-custom-adetailer-model to make this model.
FAQ
Comments (6)
Could you train it to detect feet?
I trained it mostly on the soles of feet, but there is a model out there designed specifically to detect feet in general (but it didn't work for soles). It was the base of this one.
@ReignShad0 What's the foot model?
Dumb question, but would this detect extra toes? I want to fix feet with 6 toes, and was hoping this could fix it.
Yes... and no. Actually that was going to be the next training direction of this model. It should still detect a few extra toes (according to my testing), but any more and it won't detect it as a foot anymore. Will probably need to generate a bunch of malformed feet from various models as a baseline.