search

SamSamhuns / face_registration_and_recognition_milvus

Face registration and recognition system built with Docker, Uvicron+Fastapi, Milvus, Redis, and mariadb-mysql
Apache License 2.0
27 stars 6 forks source link

Change model for detecting #4

Open aestetor opened 11 months ago

aestetor commented 11 months ago

I wanted to use your project for school project by running it on local server by FastApi/Uvicorn. But i faced problem: I ran your code, uploaded faces to milvus, but when i try to recognize face with another image of the same person, it always shows "there is no similar faces". I thought maybe issue is in model you using like "facenet-trtserver" and "face-reidentification-retail-0095".

Maybe it is real to change it yolo or mtcnn or rcnn models i want to use.

Maybe you can help me with this part?

SamSamhuns commented 11 months ago

Hi, you can try reducing the detection threshold in app_docker_compose/app/models/model.py. Otherwise, you can try a different model which might take more work to integrate unless they are already in the Triton server runnable format.

aestetor commented 11 months ago

Hi, you can try reducing the detection threshold in app_docker_compose/app/models/model.py. Otherwise, you can try a different model which might take more work to integrate unless they are already in the Triton server runnable format.

I played a bit with threshold in InputModel, but it is not working with that how expected. I already registered face, so i tried to recognize the same face in another view, but it still now work. Maybe you have some advices to change or something?

aestetor commented 11 months ago

Can you check please the models in link, are they compatible with your code?

SamSamhuns commented 11 months ago

Hi can you send those faces you are using?

aestetor commented 11 months ago

Hi can you send those faces you are using?

photos.zip

SamSamhuns commented 11 months ago

Hi, I have been very busy with other stuff but you need a model in the savedmodel, h5 or onnx format to load in trt server. Or do you have other better face recognition models?