Researchers have introduced Delineate Anything v2, a new foundation model specifically designed for mapping agricultural field boundaries on a global scale. This model addresses limitations of existing vision foundation models in geospatial applications by incorporating physical scale awareness and handling complex topological issues. To support this, a large dataset named FBIS-73M, comprising 73 million instances across 61 countries, was created, along with a new benchmark for evaluating zero-shot generalization. Delineate Anything v2 demonstrates superior performance compared to previous state-of-the-art methods and is capable of rapid, large-scale deployment, as evidenced by its ability to map Ukraine's agricultural fields in under six hours. AI
IMPACT Enhances global food security and carbon accounting through improved agricultural field mapping capabilities.
RANK_REASON Release of a new research model and dataset with performance benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
