deepface-api

This is a simple project to detect human faces in an image using deepface and fastapi.You can use it as a face detection api.

Reference

• DeepFace: A Lightweight Face Recognition and Facial Attribute Analysis (Age, Gender, Emotion and Race) Library for Python.

Requirements

• Python 3.10+
• Debian / MacOS

I've only tested this on MacOS and Debian, so its compatibility with other platforms is uncertain. Therefore, I strongly recommend using Docker to run this project.

Install

Docker (Recommended)

docker compose up -d

The docker image will be built and run automatically. Default port is 8008.

Due to the size of dependencies, it may take a while to build.

Manual

pip3 install fastapi
pip3 install "uvicorn[standard]"
pip3 install python-multipart
pip3 install numpy
pip3 install opencv-python-headless
pip3 install retinaface==1.1.1
pip3 install deepface
pip3 install tf-keras
# After installing the dependencies, you can run the api by executing the following command:
python3 api.py

API

Detect and return image file

Request

curl --location 'http://127.0.0.1:8008/detect_and_return' \
--form 'file=@"/Users/luolei/Desktop/human-image-example.jpg"'

Response

The response will return a image file with faces detected.

Detect Image and return json

Request

curl --location 'http://127.0.0.1:8008/analyze' \
--form 'file=@"/Users/luolei/Desktop/human-image-example.jpg"'

Response

The response will be a json like below:

{
  "status": "success",
  "faces_detected": 1,
  "faces": [
    {
      "position": {
        "x": 943,
        "y": 1126,
        "w": 131,
        "h": 178
      },
      "confidence": 0.9973875284194946,
      "age": 29,
      "gender": "Woman",
      "dominant_race": "asian",
      "dominant_emotion": "neutral",
      "emotion": {
        "angry": 2.710123546421528,
        "disgust": 7.478310748432193e-8,
        "fear": 2.3377612233161926,
        "happy": 0.0003928863861801801,
        "sad": 12.757988274097443,
        "surprise": 0.0003206779865649878,
        "neutral": 82.1934163570404
      },
      "race": {
        "asian": 99.99648332595825,
        "indian": 0.0013003231288166717,
        "black": 3.5335065429187296e-8,
        "white": 5.535627565222967e-5,
        "middle eastern": 7.717589761702115e-9,
        "latino hispanic": 0.002161472184525337
      }
    }
  ],
  "output_file": "/Users/luolei/Desktop/result_20241105_155522.jpg"
}

License

This project is licensed under the MIT License - see the LICENSE file for details.