Deemer
A workflow for turning expert reviews of ranch footage into training labels.
Deemer grew out of Openrange. Ranch footage can show an animal moving, but it takes experience to know which movements matter. We needed a way to bring that judgment into the data used to train a model.
I’m developing a workflow where a team uploads footage and describes what it needs to understand. Software prepares clips and questions; people with relevant experience review them from their phones. Their answers become verified labels for the team training the model.
The first use case is ranch footage. The broader idea is to make it practical for people with expertise in the physical world to help teach machines what they know.