charleneleong-ai / mrcnn-docker

MaskRCNN ML Flask App

---

Running locally

1. Clone repo with submodules

   $ git clone --recurse-submodules [e-mail]:charyeezy/mrcnn-docker.git

2. Set up Conda or virtual environment and install reqs for frontend and backend

   $ conda env create -f environment.yml
   $ conda activate mrcnn

3. Set up MaskRCNN project and pycoco tools

   $ cd backend/Mask_RCNN && python3 setup.py install
   $ cd coco/PythonAPI && make

4. Download Coco weights

    $ wget -O backend/mask_rcnn_coco.h5 https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5

5. Install Jupyter kernel and make sure mrcnn-predict runs with this kernel

   python -m ipykernel install --user --name mrcnn 

6. Jupyter Gateway

   $ jupyter kernelgateway --KernelGatewayApp.api='kernel_gateway.notebook_http' --KernelGatewayApp.ip=0.0.0.0 --KernelGatewayApp.port=9090 --KernelGatewayApp.seed_uri=mrcnn-predict.ipynb --KernelGatewayApp.allow_origin='*'

Running Flask in HTTPS

1. Use self-signed certificates using openssl

   $ cd frontend/src && openssl req -x509 -newkey rsa:4096 -nodes -out cert.pem -keyout key.pem -days 365

Build DockerFiles

1. Create data volume

   $ docker volume create --name mrcnn-data

2. Create a user-defined bridge using built-in bridge network driver for app

   $ docker network create mrcnn-net

3. Build and run frontend docker with data volume

   $ docker build --rm -f "frontend/DockerFile" -t mrcnn-frontend:latest "frontend"
   $ docker run --network mrcnn-net -itd --rm --name mrcnn-frontend -p 5000:5000  -v  mrcnn-data:/app mrcnn-frontend 

4. Build and run backend docker with data volume and connect

   $ docker build --rm -f "backend/DockerFile" -t mrcnn-backend:latest "backend"
   $ docker run --network mrcnn-net -it --rm --name mrcnn-backend -p 9001:8888  -p 9090:9090 --volumes-from mrcnn-frontend mrcnn-backend 

5. [Optional] Run with Jupyter notebook to edit

   $ docker run --network mrcnn-net -it --rm --name mrcnn-backend -p 9001:8888  -p 9090:9090 --volumes-from mrcnn-frontend mrcnn-backend  jupyter notebook --allow-root 

Frontend

1. Upload pictures using Flask-Dropzone
2. Shows prediction returned from model

Backend

Matterport's MaskRCNN is main model. The HTTP REST API is hosted using a Jupter Kernel Gateway in Jupyter notebook.

Get Docker IP

$ docker network inspect mrcnn-net
$ docker inspect <containerNameOrId> | grep '"IPAddress"' | head -n 1