Skip to content

Latest commit

 

History

History
 
 

face-recog

README.md

keras-serving face-recog (emotion and gender detection)

Info

  • models.py, utils.py and test.py are taken (and adjusted) from oarriaga
  • I am using the pre-trained models from oarriage, which he trained on the fer2013.bib dataset
  • Image face-classification is done by opencv's haarcascade
  • oarriaga's project is under MIT License

1. Install Requirements

NOTICE: call these from the project root, not from inside the ./face-recog directory

./face-recog/prepare.sh
# installs python dependencies via pip
# also make sure you already have built the docker-image for the XOR project /README.md#1

2. Re-import and export Model as Tensorflow Graph

python face-recog/export.py # make sure to run from project root folder
# results will be in (/export)

3. Build & Run Containers via docker-compose

./start-dual-servers.sh
# ./stop-dual-servers.sh

4. Test API via client

  • the client loads an image from whatever path (relative/absolute) you pass as first argument
cd ./face-recog/client
npm install
npm start ./../images/cutouts/4.png
# make sure to start the servers first

5. Generate and Test on your own images

cd ./face-recog/
python face-exporter.py ./images/chris_1.jpg 
# makes a gray cutout in 48x48 pixel of a face in the image
cd client
npm start ./../images/face-exports/gray_0.png