Doodler

We achieve a fully reproducible workflow by using a semi-interactive ML program called ‘Doodler’ that uses sparse annotations contributed by human labeler to estimate classes for all pixels. Each label image may be reconstructed using the sparse annotations provided by a human annotator, and further, those annotations might be repurposed using a different algorithm, if necessary.

To get started with Doodler, go to the Doodler website and read the paper

The code is found on github. There is also an official USGS software release for Doodler.