SegMate Toolkit

SegMate is an open source AI Toolkit developed by the Vector Institute, which can help organizations and researchers apply cutting-edge computer vision techniques in the fight against climate change. The toolkit includes an intuitive interface to apply Meta’s Segment Anything Model (SAM), allowing users to precisely isolate individual objects in an image, an important step in recognizing, classifying, and segmenting objects in machine learning applications. 

The model, created in collaboration with BMO and TELUS, as well as other partners, can be applied to satellite images to track changes in deforestation, agricultural land use, ocean and inshore water bodies, and natural disaster responses. It replaces current methods, which rely on hand annotation and analysis, and can help companies better understand potential environmental risks to their investments.

SegMate extends SAM’s field of view from strictly natural objects to accommodate satellite imagery and allows researchers to infer high-resolution regions from the satellite imagery inputs, like forest cover across large areas, a key factor in understanding the human impact on forests. It also automates region of interest generation at the scales demanded by remote sensing applications and provides a prompt-based method for defining the desired masking using natural language.  

The initiative underscores the value of co-development practices when bridging the gap between state-of-the-art research and real-world application. By emphasizing open-source solutions, the partners hope to increase the potential for widespread application, offering promising pathways for future advancements in the field.