This is a server backend for face classification to be used via API from e.g. a Hololens client.

• Python 3
• Flask
• Tensorflow
• OpenCV
• Numpy
• Scikit-learn
• Scipy
• Requests (Python package)

• git clone https://username@bitbucket.zuehlke.com/scm/daana/hololens_backend.git path/to/local
• (git checkout development check out development branch to begin development work!)
• (Use default checked out master branch for deployment)

• python3 server.py
• This will take a while, since face detection and classification models are loaded
• When Flask reports Running on http://0.0.0.0:5000/ (Press CTRL+C to quit) the server is ready to receive images

• python3 client.py
• This will load a test image, serialize it, send it to the server via API, and print the response
• (This should print out {'bounding_boxes': [[247, 132, 472, 412]], 'predictions': ['others']} for the face.jpg test image, i.e. a face is found, but it could not be recognized)

• master branch is the stable branch and should always be in a working and stabel state
• development branch is for current development
• Use feature branches to test things out and implement additional features, then merge into development for integration/testing, then merge into master for deployment

• server.py implements the server in Flask
• configuration.py defines various settings, server port, model directories, logging behaviour
• logger.py implements an event and error logger
• classifierFace.py implements face detection and face classification
• client.py implements an example client to communicate with the server
• model/ contains face detection and classification models used by the face classifier
• log/ contains event and error logs, as well as logged image files
• test/ contains test files

• In the example Python client to Python server communication, the image is serialized to a byte string using Python's pickle, and unpickled on the server side.
• To customize how images are sent from a client to the server, the corresponding sections are marked both in client.py as well as in server.py (it is important that serialization and deserialization match exactly, such that the image is understood by the face classification module.)
• Ensure that MD5 hashes of image data on client and server are equal (see code)

• Flask runs in non-threaded mode, i.e. requests are processed one-by-one
• Debug mode is turned off such that the server accepts connections from outside localhost

• The server supports some logging, mainly event logging and error logging, using the logger.py module
• The server can also write every single image it receives to disk, with a timestamp

• TODO

• Install XCode
• Install OpenCV
  • brew tap homebrew/science
  • brew install opencv3 --without-python --with-python3 --with-contrib
  • echo /usr/local/opt/opencv3/lib/python3.6/site-packages >> /usr/local/lib/python3.6/site-packages/opencv3.pth
  • (In case of development in a virtual environment) echo /usr/local/opt/opencv3/lib/python3.6/site-packages >> ./env/lib/python3.6/site-packages/opencv3.pth
• Install Python dependencies
• (Optional: set up virtual environment)

• TODO

• URL: http://ip:port/api/detect/faces
• Purpose: detect faces in an image and respond with face locations as well as face classificationsmodels
• State: implemented
• Request: see client.py

• URL: http://ip:port/api/admin/reload
• Purpose: send a request to the server in order to reload models
• State: template for request handler implemented
• Request: curl --header "Content-Type: application/json" -X POST --data '{"command":"1"}' http://localhost:5000/api/admin/reload