PLANTEUR Frozen - Inputs Pixel Accuracy - S1+S2 - Readout 64d: leaderboard
PLANTEUR Frozen - Inputs Pixel Accuracy - S1+S2 - Readout 64d: Accuracy (%). PLANTEUR Ai2 October22 2026 technical report: frozen encoders and released annual embedding products only. Probe heads are fitted evaluation readouts; backbone weights never updated. OlmoEarth v1.3 is one d768 encoder with 64d/128d distilled projections, kept as readout configurations. Supervised scratch and fine-tuned encoder rows, including PASTIS-fine-tuned encoders in Table10, are held. Main test443 windows, eight classes, void excluded, absent LOIO classes score0. Probe learning rate selected on validation among8 candidates, 50epochs. Native sensors/windows/label regimes and mean-IoU vs pixel accuracy remain distinct. Numeric PDF bytes pinned in full; no model inference. Native source condition: Readout 64d.
Metric: Accuracy (%). Source: allenai.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | OlmoEarth v1.3 | 90.21 |
Interactive version: theaggregate.ai/benchmark?slug=planteur-frozen-inputs-pixel-accuracy-s1-pluss2-readout-64d · How It Works · Data refreshed daily, snapshot 2026-11-12.