README.md

Face Landmark Detection

This API is based on the work found here.

It’s designed to be used with Request Directory and you can find more details here

Development

Install dependencies

pip install -r requirements.txt

Run the API

python src/main.py

Usage

By default, the API runs on port 7002.

docker run -it -p7002:7002 ghcr.io/arsenstorm/facelandmarks:latest

API

To use the API, you need to send a POST request containing form data to the /landmarks endpoint with the following parameters:

Parameters

• image: The image to detect landmarks on.

Example Request

As an example, we’ll upload this image:

curl -X POST http://localhost:7002/landmarks -F "[email protected]/facelandmarks/example_input.jpg"

Example Response

We get the following response:

{
  "bounds": {
    "x1": 339,
    "x2": 670,
    "y1": 188,
    "y2": 605
  },
  "confidence": 0.9646592140197754,
  "face": {
    "left_eye": {
      "x": 425,
      "y": 375
    },
    "left_mouth": {
      "x": 476,
      "y": 494
    },
    "nose": {
      "x": 518,
      "y": 412
    },
    "right_eye": {
      "x": 560,
      "y": 315
    },
    "right_mouth": {
      "x": 615,
      "y": 438
    }
  },
  "image": "/9j/4AAQSkZJRgA...", // shortened for brevity
  "landmarks": [
    {
      "x": 362,
      "y": 395
    },
    // ...
    {
      "x": 550,
      "y": 324
    }
  ],
  "success": true
}

A total of 98 landmarks are returned, alongside the face bounds and the confidence of the detection.

The image is returned as a base64 encoded string, which can be decoded where it’ll look like this:

Tip

You can test out the API using Request Directory without needing to run it locally.

Notes

• Testing indicates that the API needs roughly 2GB of memory to run most optimally.
• When no API requests are running, the memory usage is roughly in the range of 0.8-1.0GB.