Make a few doodles ... classify entire scenes

Identify pixels of each of the 'classes' present in the scene with a mouse/stylus, and it does the rest, using a model to 'auto-complete' all the pixels you didn't label.

A class is a discrete label such as 'sky', 'water', 'sand', etc. Auto-completion refers to the process of classifying each pixel in the scene into a class using an automated process, otherwise known as image 'segmentation'.

For images of natural environments

You can use any type of photos, but it is designed to work best with imagery consisting of landscapes (natural environments), where the surface composition, cover, and sometimes evidence of human and other animal uses, are identifiable as characteristic textures and colors.

Doodler uses Machine Learning to classify each pixel of the scene, i.e. segment the image, which is optimized for classifying such natural textures and colors, which can vary considerably for any given class. Often this 'model' needs most help (i.e. more doodles) near the boundaries where one class transitions to another.

Generate high volumes of labeled imagery quickly

There are many great tools for exhaustive (i.e. whole image) image labeling using polygons, such as makesense.ai. However, for high-resolution imagery with large spatial footprints depicting complex natural scenes, such as imagery collected from airplanes or satellites, exhaustive labeling using polygonal tools can be very time-consuming and, well, exhausting!

Doodler is as an alternative tool for rapid approximate segmentation of images that is semi-, not fully, supervised by you.