Finished! Looks like this project is out of data at the moment!
We're happy to announce that our preprint is out, describing our tree mapping research and the (now public) dataset that this project uses! You can find it on arXiv here and check out the models for yourself here. Thanks again for volunteering, your efforts are helping us improve our labels and ultimately improve the state of tree and restoration monitoring.
We've completed the classification stage for this project! To browse other active projects that still need your classifications, check out zooniverse.org/projects.
FAQ
What will we do with the data from this project?
We will use the results from Zooniverse to help validate our labelled data.
Most of the images you will be shown have already been labelled by a single human. We label images with two classes: single trees and multiple/groups of trees. We think our dataset is generally high quality - most of the labels are correct, and we have high quality masks around tree cover - but it’s time consuming to check each image individually to try and find mistakes.
By looking at where citizen scientists label trees, we can check if our dataset agrees. Where there is disagreement, for example only 1/10 people say there is a tree where we have a label, then we can flag it for review.
Where does the data come from?
Our data is primarily sourced from Open Aerial Map, an amazing repository of user-contributed aerial surveys. These images are high resolution and are georeferenced so we know exactly where they were captured.
In this project, we have also included data from conservation projects monitored by Justdiggit in Tanzania, some imagery from restoration partners and some imagery from DeepForest, one of the most popular existing deep learning-based models for tree detection.
What is the background image?
The background image for the project was taken from Wikimedia, it is an established Beech forest in Belgium: https://commons.wikimedia.org/wiki/File:Brussels_Zonienwoud.jpg
Who provided translations?
We would like to thank colleagues at Restor for localizing the project into several languages. Also special thanks to Louis (@veragon) who provided the French translation!