This repository contains a web service that finds and classifies symbols of music notation. The service works .png and .jpg images that are scaled to a staff interline of 10pt.

The service is based on the microframework Flask. The main route is /classify, it lets you upload an image (POST to image). The image will be scanned using a Deep Neural network and the bounding boxes of the detected symbols are returned in a json formatted string.

It is built to be used in conjunction with trained models following the DeepWatershed-Detection architecture. see: https://github.com/tuggeluk/DeepWatershedDetection

• main.py contains the flask app and calls the detection system located in class_utils.
• class_utils/ contains the code that loads a pretrained tensorflow model and performs the detection.
• demo/ contains a test/demo script for the service.
• important: for this service to work you need to create a folder called trained_models and place the correct pretrained models in said folder. (can be found at: https://drive.google.com/open?id=1Knm26FjS6YMrBVU009mRTc19ims1oj6R)
• Dockerfile The dockerized version of all of this to reduce the amount of maintanence and administration necessary

We assume you're in the home directory.

1. Build the docker image:

docker build dockerized_classifier/Detection_Service -t saty/detection

The -t argument tags the image with the label saty/detection

2. Create a self-restarting container from the image

docker create --restart unless-stopped --gpus all -p 5000:5000 --name detection_docker saty/detection

3. Start and enable the service associated with running the docker container

sudo service detection-docker start
sudo systemctl enable detection-docker  # To auto-start the service on reboot

4. To check the status of the service

service detection-docker status

5. To turn of the service

sudo service detection-docker stop
sudo systemctl disable detection-docker  # Only if you want to prevent the service from restarting on reboot