face-annotator-core-ml

A tool that detects faces and automatically annotates the size and coordinates of the bounding box. The tool uses BlazeFace to detect faces. The generated JSON file conforms to the format required by Core ML object detection, as seen below.

[
  {
    "image": "image1.jpg",
    "annotations": [
      {
        "label": "Anna",
        "coordinates": {
          "x": 120,
          "y": 164,
          "width": 230,
          "height": 119
        }
      },
      {
        "label": "John",
        "coordinates": {
          "x": 230,
          "y": 321,
          "width": 50,
          "height": 50
        }
      }
    ]
  },
  ...
] 

Requirements

• Node.js

Installation & Usage

1. Clone the repo and run npm install.
2. Run the app: node app.js /path/to/images label. The generated JSON file will be placed in the same folder as your images. If a file named annotations.json already exists in the folder, the script will ask whether you want to append the annotations in that file with the annotations you just created.

Notes

For compatibility, this tool uses the base @tensorflow/tfjs package. For increased performance, install @tensorflow/tfjs-node or @tensorflow/tfjs-node-gpu. Both of these require Python to be installed (and CUDA for the latter). You should also change the first line in annotate.js to require one of these packages.