slavabarkov / tidy

Offline semantic Text-to-Image and Image-to-Image search on Android powered by quantized state-of-the-art vision-language pretrained CLIP model and ONNX Runtime inference engine
GNU General Public License v3.0
229 stars 15 forks source link

Face recognition #8

Open test2a opened 1 year ago

test2a commented 1 year ago

Would it be possible to tag faces and search for them?

slavabarkov commented 1 year ago

@test2a It's a cool suggestion, but I am not sure about the tagging feature yet. I will consider it, but probably not anytime soon. Meanwhile, image to image search that is already implemented sort of solves this task, at list partially.

test2a commented 1 year ago

@slavabarkov i checked the image to image search but that did not show good results, as you said partially. it appears to me the image-to-image search is looking for a person mostly as opposed to recognizing facial features.

can we have the model try to recognize faces and then specifically index facial features so that they can be recognized from other people.

alnmy commented 1 year ago

Would it be possible to do facial recognition with CLIP? I guess it might be somewhat possible but I feel that would require a more specialised model, though disclaimer I know like nothing about this.

test2a commented 1 year ago

@alnmy is there a model that is designed for faces specifically?

alnmy commented 1 year ago

@alnmy is there a model that is designed for faces specifically?

While I have 0 experience in AI, I know that PhotoPrism has facial recognition which they explain here: https://docs.photoprism.app/developer-guide/metadata/face-recognition/