Requirements :

Python(>=3.5) pip(>=9.0.1)

Architecture

Install

sudo apt-get update
source install.sh [SERVERNAME]

You can safely ignore the red lines...

SERVERNAME can be provided as an argument or it will be asked during installation. (examples : "localhost" or "http://my_url.com http://www.my_url.com" or "143.21.178.46") Keep in mind that the address/name should be accessible from outside, otherwise you will experiment some 404 errors...

Installation is made in a Python VirtualEnv so you must use the "activate" command to manually activate the Env (mandatory to DEBUG/TEST but automaticaly done by supervisor while in production)

Run API:

#DEBUG DETECTOR (DEBUG flag activated by default)
source ./env/bin/activate
python3 reco_label.py IMG_URL [CONFIDENCE]

#DEBUG FLASK
source ./env/bin/activate
gunicorn api:app

#PRODUCTION (auto start at boot)
sudo service supervisor start
sudo service supervisor restart
sudo service supervisor stop
sudo service supervisor status

with IMG_URL : an accessible image url CONFIDENCE : the confidence threshold (def to 65, mean that detection are drawn over 65% confidence only)

Scale

Add --workers (set the number of workes instance) option to the detect_label.conf file, at the end of "command" line like following:

command=[USERNAME]/env/bin/gunicorn api:app -b localhost:8000 --workers=42

(before installing, in the surrent directory or after installed, in /etc/gunicorn/conf.d/detect_label.conf, then restart supervisor)

Diagnostic

sudo service nginx status
sudo service supervisor status
cat /var/log/detect_label/detect_label.err.log

Results image and data can be accessed by urls. You can access to the results directory by editing nginx config file as following:

location /results/ {
        autoindex on; 

on instead of off (default). Then restart nginx (sudo service nginx restart) and go to http://[SERVER_NAME]/results/

Run UnitTests:

python3 -m pytest --disable-warnings

Authorize ip

(https://codepen.io/corenominal/pen/rxOmMJ)

echo "API_KEY:IP" >> auths.txt

with API_KEY : your private API_KEY IP : your public IP Several IPs can be user wuth the same API_KEY, just add one line per IP You can use wildcards "*" at the end of the IP to allow a whole network/subnet

Call api (IP must have been authorized)

curl http://[SERVERNAME]/api/?key=API_KEY&source=IMG_URL&confidence=CONFIDENCE&debug=DEBUG&fileout=FILEOUT

with API_KEY : your private API_KEY (note that it is tied to your IP by host) IMG_URL : an accessible image url CONFIDENCE : the confidence threshold (def to 65, mean that detection are drawn over 65% confidence only) FILEOUT : flag (0 or 1) to get the censored image url in return (default to 1) DEBUG : debug flag (0 or 1) save and return original image, masked image, boxes data and detected image (default to 0)

returns : { "message":"OK", "result_data": {Boxes data dict}, ["result_image":"your_image_name_censored.jpg"] }

Remove service

source uninstall.sh