This is a simple tool for generating an interface to quickly build training sets to feed into deep learning computer vision models.

We've been able to tag batches of 1000 photos in about 30 minutes. It could be faster or slower for your use case depending on the complexity of your classification problem.

Since the interface let's you manually classify images so quickly, it's often easier/faster to generate training sets yourself rather than taking the time to train others and/or integrate with a platform like Mechanical Turk.

python tagging_html_generator.py \
  --photos-file example-local-photos.json \
  --config-file example-local-config.json \
  > ./build/index.html

python -m http.server

Then open http://localhost:8000/build/index.html and tag away.

Copy the console output after hitting Enter into your favorite file editor to clean up your results for use in training. Yeah, really.

Simply set formAction in your config file to the External Question Form Action and generate the html just as we did above.

You can then insert the generated html into an HTMLQuestion.

There are two parts to this html generator:

1. The javascript
2. The python html generator

To update the distributed javascript and css in ./dist/, run the following:

pushd react-ui && \
  npm install && \
  npm run-script build && \
  popd

The javascript is currently not tested, but you can run tests for the html generation with the following:

python -m unittest tagging_html_generator_test