Cord

After (too much) time spent hacking, we finally launched the Web App, one of three core components of our training data labelling interface.

It allows you to:

• annotate, classify, and segment images and video
• visualise data
• perform collaborative QA
• train labelling models with the click of a button

Labelling for computer vision in order to build kick-ass models is especially arduous, so we have put a lot of care into building a tool that can very efficiently label video and images using a collection of in-built heuristics and embedded machine learning models, what we call “labelling algos.”

It took us a while, but our toolbox has gotten to 16x labelling efficiency vs. widely used alternatives (https://www.medrxiv.org/content/10.1101/2020.09.11.20192500v1.full.pdf)