Authors: Jenny Huang, Ian Hunt-Isaak, William Palmer

Write up

This was our final project for https://harvard-iacs.github.io/2020-AC295/. We wrote a medium article about this! Check it out here: https://medium.com/institute-for-applied-computational-science/how-we-built-an-easy-to-use-image-segmentation-tool-with-transfer-learning-546efb6ae98

Image segmentation code is now availiable on PyPi!

The image segmenter in this project is now availiable on pypi via mpl-interactions, see documentation here. Also this project inspired ipysegment, an HTML5 canvas based canvas based image segmenter for use in jupyter notebooks. ipysegment will be significantly less laggy to use, but currently has fewer features.

Trying it out

docker

From Dockerhub:

sudo docker run -p 8888:8888 -e JUPYTER_ENABLE_LAB=yes ianhuntisaak/ac295-final-project:v3

Binder
We don't recommend this, everything will be very slow (also it currently fails to build 😢):

Installing dependencies

Or you could install it locally. We have a conda environment file in the binder directory. So you can make an environment with the following commands:

conda create -n segmentation -f binder/environment.yml -y && conda activate segmentation
jupyter labextension install @jupyter-widgets/jupyterlab-manager --no-build
jupyter labextension install @jupyter-widgets/jupyterlab-sidecar jupyter-matplotlib

Import structure

There are separate directories for notebooks and python files because otherwise things become a real mess. To import something from a python file in the lib directory put the following at the top of your python notebook:

# set up path for relative imports
import os
import sys
module_path = os.path.abspath(os.path.join('../'))
if module_path not in sys.path:
    sys.path.append(module_path)

Then you can do from lib.____ import ____

Updating docker image

Follow: https://ropenscilabs.github.io/r-docker-tutorial/04-Dockerhub.html

build

sudo docker build -t ianhuntisaak/ac295-final-project:<tag> .

Verify that it works If you use a port other than 8888, e.g. -p 8889:8888 then you need to change the port in the URL printed in the terminal, can't just copy paste.

sudo docker run -p 8888:8888 -e JUPYTER_ENABLE_LAB=yes ianhuntisaak/ac295-final-project:<tag>

push to dockerhub

sudo docker push ianhuntisaak/ac295-final-project

update the tag version in this readme Turns out using the latest tag is a bit of a nightmare: https://medium.com/@mccode/the-misunderstood-docker-tag-latest-af3babfd6375