Case Study: Omdena achieves rapid, high-accuracy tree identification for wildfire prevention with Labelbox

Managing a Labeling Team and Building a High-Impact Deep Learning Model for Tree Identification

Omdena ran a global AI challenge to build a satellite-imagery tree identification model to help prevent fires and save lives. Leo Sanchez led a 15-person labeling team within the Omdena project and the team used Labelbox for image annotation and workflow management, with Labelbox providing a free educational license so all contributors could work in a single centralized account.

Labelbox’s polygon and box annotation tools, queueing, review and export features let the team label the dataset quickly and consistently, enabling high-quality training data for model development. Using those labels, the team trained a Deep U-Net that reached ~94% validation accuracy in initial runs and 95% after data augmentation, clearly separating trees from shadows on unseen images and supporting Spacept’s wildfire-prevention efforts—outcomes made possible by Labelbox’s centralized labeling and QA capabilities.

Omdena