Navigation :

Web Service

happy-day-service

With make all the scripts also starts an instance of the created docker container. It listens on port 80, so keep in mind that you need the rights to listen on this port. Otherwise change the port in the Makefile.

# clones happy-day-service repository
git clone git@github.com:ofesseler/happy-day-service.git

# builds new containers from scratch
make all

If you want to be able to store the images on a webdav endpoint, you have to define the environment variables DAV_USER and DAV_PASSWORD. Otherwise you can’t store data for retraining and collecting without issues.

Requirements

Python 3.x
Tensorflow 1.4
Keras 2.x
pip
docker (optional)

For further requirements take a look into requirements.txt

Training

To train a new model, one can easily download the Dataset from a Kaggle Challenge “Facial Expression Recognition”. The other option is to generate the data yourself and take a lot of photos from your friends. We had about 5000 photos of 15-20 Friends and it helped a lot to improve the modell.

Do the following steps to start training the self-cnn model

# change into happyday folder
cd happy-day-service/happyday

# preprocess data for scaling, folder structure, ... 
python data_processing.py -s "/source/path" -d "/destination/path"

# train model 
python self_nn.py --path "/destination/path"

Then your machine will work a while and eventually show you the results. You can modify the parameters in the self_cnn.py file itself.