Python Face Serving

Face recognition serving APIs based on FastAPI and Onnxruntime.

License

AGPL 3.0
Copyright © 2024 Hieu Pham. All rights reserved.

Prequistance

!wget https://huggingface.co/datasets/hero-nq1310/stuffhub/resolve/main/pyfaceserve-models.zip
!unzip pyfaceserve-model.zip

One step with docker compose

docker-compose up

Customizaition in doker-compose.yaml

In triton service, you can run with cpu/gpus by commenting/uncommenting these following lines:

deploy:
    resources:
        reservations:
          devices:
          - driver: nvidia
            device_ids: ['1']
            capabilities: [gpu]

Change environment variables in pyfaceserve service when need to change threshold, model_name, API:

• TRITON_URL=localhost:6000
• DETECTION_NAME=yolov7-hf-v1
• SPOOFING_NAME=spoofer
• RECOGNITION_NAME=ghostfacenet
• DETECTION_THRESH=0.7
• SPOOFING_THRESH=0.4
• RECOGNITION_THRESH=0.4
• QDRANT_URL=localhost:6333
• IMG_DIR=face_images

How to scale with different real case

Create new docker image with new value of DB_NAME and volumes in docker-compose.yaml file.

    image: heronq02/pyfaceserve:v1.0.0
    environment:
      ...
      - IMG_DIR=<new_path>
      - DB_NAME=<new-name-collection>
    volumes:
      - ${PWD}/<new_path>:/<new_path>
    ...
    command: fastapi run main.py --port <new_port>

DB_NAME will set database hard-code with different port API.