Empowering Sustainability through Geospatial Image
Type of service
Through their exceptional work in geospatial image instance segmentation, particularly in the realm of tree identification, Restor is revolutionizing the way we approach environmental conservation efforts.
At Humans in the Loop, we are delighted to showcase our successful partnership with Restor, a company that is making significant strides in promoting sustainability.
Creating a world where every action positively impacts nature. Restor is an open-data, geospatial platform that offers nature stewards access to the latest ecological science for impact storytelling. With 130,000+ sites, Restor is the largest network of community led restoration and conservation sites across the globe. Through Restor, governments, companies, impact funders and financial institutions can connect and contribute to the regenerative use of the land we all depend on.
Restor, as a leader in sustainable initiatives, recognized the importance of accurately classifying trees and canopies within geospatial images to better understand and protect natural ecosystems. They approached Humans in the Loop with a challenge – to annotate and segment more than 5,500 geospatial images containing a wide variety of tree sizes and complexities.
One of the key challenges encountered during the annotation process was the significant variation in the number of objects and the complexity levels within each image. Some images required intricate segmentation, particularly when dealing with large canopy masks, while others involved well-separated circular crowns that were relatively straightforward.
To ensure the highest quality of data, Restor adopted a stringent approach, flagging ambiguous or edge case images to prioritize cleaner data. This approach aimed to avoid any potential confusion during the training process.
Working with Humans in the Loop was a great experience, especially considering how nuanced and intricate our annotation protocol is. I was very happy to see how hands-on and proactive the team at HITL was, keeping in constant real-time communication on edge cases, feedback from our side, and changes to the annotation guidelines. This attitude and process really made for a successful annotation sprint and I’m looking forward to continuing our relationship.
Dr Josh Veitch-Michaelis, Researcher in AI for Remote Sensing, Restor
Humans in the Loop worked closely with Restor to develop a customized annotation strategy, using individual polygons to precisely label and segment trees. The annotation was performed according to the COCO (Common Objects in Context) format, ensuring compatibility and ease of integration into Restor’s existing systems.
Special attention was given to flagging ambiguous or edge case images, allowing for a clear delineation of the training data and ensuring the accuracy and reliability of the annotation process.
Restor can now employ the annotated results to develop advanced algorithms, enabling them to effectively identify and monitor individual trees and canopies within their projects. This enhanced understanding of forested areas will empower Restor in making environmentally conscious decisions and implementing targeted conservation strategies.
Contribute to scaling up the protection and restoration of forests around the world by labeling trees. Join Restor’s citizen campaign on Zooniverse.