PLANTEUR benchmark maps crops across French overseas territories

PLANTEUR, posted to arXiv on November 3, is a crop-type segmentation benchmark for the French overseas territories, covering six tropical regions, eight crop classes, and three aligned satellite data streams.

The authors state the setup plainly. "We introduce PLANTEUR, a benchmark for crop-type mapping from satellite images in the French overseas territories," they write. "Building on the PASTIS design, PLANTEUR provides georeferenced and aligned Sentinel-2, Sentinel-1, and Landsat-8 sequences with parcel-derived labels spanning six tropical regions and an 8-class label space."

Two protocols are defined: a standard test split, and a leave-one-island-out setup built to measure cross-territory generalization. Evaluation covers fully supervised models, linear probes, and fine-tuned foundation models. The authors say the benchmark code and data are open source on GitHub and Hugging Face.

The abstract reports no per-model accuracies, so which approach wins the island-out protocol is a question the paper body has to answer. Two researchers we follow flagged the paper within a day of its posting.