foivospar / Arc2Face

[ECCV 2024 Oral🔥] Arc2Face: A Foundation Model for ID-Consistent Human Faces
https://arc2face.github.io/
MIT License
570 stars 39 forks source link

Using without models from insightface? #1

Open JohnMe opened 6 months ago

JohnMe commented 6 months ago

Hi, is there any way to train this and without using the models from insightface, since they have a really restrictive license...

jadechip commented 6 months ago

+1

jadechip commented 6 months ago

Also curious if you are planning on releasing the training code? Thank you.

foivospar commented 6 months ago

Please see https://github.com/foivospar/Arc2Face/issues/3#issuecomment-2023165300.

jadechip commented 6 months ago

@foivospar could you kindly elaborate a bit on how insightface is being used, and if it's non-commercial license applies to arcface? From my understanding of the antelope package,

scrfd_10g_bnkps.onnx is used for initial face detection.
2d106det.onnx provides detailed facial keypoints.
glintr100.onnx is responsible for generating face embeddings.

...does arcface replace these?

foivospar commented 6 months ago

No, we only replace the face recognition module, i.e. glintr100.onnx, with arcface.onnx. The rest are downloaded from InsightFace, so their license applies. By the way, only the detection module scrfd_10g_bnkps.onnx is needed, as we do not use keypoints, gender annotations, etc.

jadechip commented 6 months ago

Thank you @foivospar. I am looking into training a non-commercial version of the insightface models, including scrfd_10g_bnkps.onnx. Will report back if I make any progress.

fofr commented 5 months ago

@jadechip Have you made any progress on your version?