Delineate Anything v2 maps field boundaries across 61 countries with a 73M-instance dataset
Mykola Lavreniuk · hf · 2026-07-22
Delineate Anything v2 is a global foundation model for field boundary mapping
This paper introduces Delineate Anything v2, a geospatial foundation model for agricultural field delineation at large scale. The authors argue that general-purpose vision models such as SAM often fail in geospatial settings because of topological complexity, crop texture patterns, and weak physical scale awareness.
Main contributions
- FBIS-73M, a 73-million-instance multi-resolution dataset spanning 61 countries
- A data-curation pipeline aimed at reducing multi-field parcel merging
- A new manually curated benchmark covering 100 countries for zero-shot generalization
Reported results
- The model beats the previous Delineate Anything framework by 0.284 [email protected], a 103.3% relative gain
- It can map all of Ukraine’s 603,000 km² in 5.4 hours on a consumer-grade workstation
Code, weights, the dataset, and national-scale vector products are publicly available.
More from Research
- Richard Hamming’s 1995 lecture asks why mathematics works at all — melnykowycz · 2026-07-22
- Meta finds quantized reasoning models often doubt the right answer instead of finishing — rohanpaul_ai · 2026-07-22
- Robot manipulation needs mobile navigation and standardized benchmarks, researchers argue — YuXiang_IRVL · 2026-07-22
- Survey maps multimodal LLMs’ weak spot: understanding memes and comics — Tuo Liang · 2026-07-22
- BigMac keeps LLM pipeline speed while capping multimodal activation memory — 小红书技术REDtech · 2026-07-22
- Four teams independently shipped the same “LLM wiki” pattern after Karpathy’s gist — garrytan · 2026-07-22