Max Hamilton

The text comes from Wikipedia articles of each species, and the observation data is from iNaturalist, same as SINR (arxiv.org/abs/2306.02564). Because of the observation data, predictions are biased toward regions enclosed by actual species boundaries, making human-made borders tricky

Combining Observational Data and Language for Species Range Estimation Species range maps (SRMs) are essential tools for research and policy-making in ecology, conservation, and environmental management. However, traditional SRMs rely on the availability of environmental...

🔗Check out the our paper and code:
Paper: arxiv.org/abs/2410.10931
Code: github.com/cvl-umass/le...

If you'll be attending #NeurIPS2024 this week, stop by our poster to learn more!
December 13th, 11am-2pm, East Exhibit Hall A-C #3903

🌎LE-SINR is able to geographically ground text prompts to locations on the earth. This includes continents and countries, geographic features, and even concepts that do not appear in our species training data but are represented in the language model 🔍Example: "hello kitty"

🚀Le-SINR combines millions of citizen science species observations with textual descriptions from Wikipedia, enabling zero-shot range estimation from text and facilitating the learning of rich spatial covariates at a global scale.

Low-dimensional projection of the learned features 👇

❓How can we predict where a species may be found when observations are limited?

✨Introducing Le-SINR: A text to range map model that can enable scientists to produce more accurate range maps with fewer observations.

Thread 🧵