Field-boundary accuracy on a 100-country manual benchmark improves from 0.275 to 0.559 mAP@0.5 (about 2x) by cleaning labels and editing training pixels rather than changing the model architecture.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Delineate Anything v2: A Global Foundation Model for Field Delineation
Field-boundary accuracy on a 100-country manual benchmark improves from 0.275 to 0.559 mAP@0.5 (about 2x) by cleaning labels and editing training pixels rather than changing the model architecture.