Our team started Digital Reclamation because we were frustrated. We wanted to work on fundamental research for recycling sorting, but there was never enough training data available to the public- all the major datasets were proprietary and labeling our own on a research budget wasn’t feasible.

We realized that the solution was right in front of us- people performing manual sorting already provide labels, if we could just find a way to make them legible to machines. Digital Reclamation represents the culmination of those efforts- a general purpose algorithm to extract pixel-wise segmentation masks from videos of any sorting activity. No retraining, no huge call centers filled with human annotators. Just provide videos of the sorting being done, labels of where each class goes, and let us return the training masks in under 24 hours.