MillionTrees: A global benchmark for tree detection across geography, sensor, and annotation geometry

MillionTrees provides over a million annotated trees and six million weakly supervised annotations for airborne RGB tree detection across diverse geographies, forest types, and annotation sources.

• Why it matters: Existing airborne tree detection datasets are constrained to single geographies, resolutions, or annotation styles, limiting generalization and making global benchmarking difficult.
• What they did: The team compiled existing and newly published annotation datasets into MillionTrees, then benchmarked multiple open-source detection algorithms using a provided evaluation standard and held-out RGB validation.
• The result: Current open-source models do not generalize across the benchmark, but training on MillionTrees improves both within-domain and out-of-domain performance, while held-out terrestrial LiDAR–paired RGB shows common test overfitting and hyperparameter tuning challenges.

AI summary of the abstract · may contain errors · not medical advice · read the paper.