By Platanus
Openface-api is a web API that compares faces on 2 given pictures, and gives back their distance (measure of unsimilarity). This is useful to check if both pictures belong to the same person. It was developed for KSEC project, but can serve general purposes.
Openface-api is a Python Flask application, with a face-comparing script based on Openface face recognition library.
In order to get started, the following must be done:
- Install Openface https://cmusatyalab.github.io/openface/setup/
- Install and setup Flask (probably with
sudo pip install Flask). Remember to set up FLASK_APP env var - Download and copy to root folder the following files: shape_predictor_68_face_landmarks.dat (extract before copying) and nn4.small2.v1.t7 (just copy)
After everything is setup, you should be able to run flask
Run this to setup you environment
virtualenv venv #only the first time
venv/bin/activate
To setup python dependencies and heroku run
bin/setup
A route is provided to check if the server is up: HTTP GET /status/
The API route is /compare/. Expects HTTP POST with the params img1 and img2 as the urls of both pictures with faces to compare. Returns a json object with the computed distance: {distance: 0.1752}
Then you can define your own distance threshold in order to evaluate if both pictures belong to the same person. Reasonable values are around 0 - 0.005. So any value higher than that means both faces are not from the same person.
Additionaly, it is possible to run the distance comparer locally in the server, executing in the terminal `python local_exe.py
To deploy on heroku we will use the beta support for docker. You will need docker to build a new deployment locally.
Install the heroku container plugin
heroku plugins:install heroku-container-registry
Login to the heroku container registry
heroku container:login
Deploy the container
heroku container:push web
heroku container:push web --remote production