TBNAI
FairRSFM: A Biome-Aware Benchmark and Debiasing Framework for Remote Sensing Foundation Models
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 04/10/2026, 20:00:00 UTC
Briefing created: 22/10/2026, 05:02:49 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 16:18:24 UTC.
Briefings
FairRSFM introduces a biome-aware benchmark to evaluate ecological group robustness in remote sensing foundation models. Using Prithvi-EO-2.0, SatMAE, and DOFA across three random seeds, the study reveals that aggregate performance consistently masks biome-dependent disparities across architectures and tasks. The benchmark covers four downstream datasets including m-EuroSAT, m-BigEarthNet, m-SA-Crop-Type, and MMEarth20K with Dynamic World label maps. Key findings include Prithvi-EO-2.0 reaching 90.98% overall macro-F1 on m-EuroSAT but a mean worst-group score of only 83.72%, and m-SA-Crop-Type dropping from 27.30% overall mIoU to 18.47% in the Xeric and Mineralogical group. Mitigation baselines like BOLP and DBR show model- and task-dependent effectiveness. Code and datasets are available at: https://github.com/aminurhossain/FairRSFM.
Key points
- FairRSFM maps georeferenced samples from 14 terrestrial biome classes into six ecologically meaningful macro-groups to evaluate ecological group robustness in RSFMs.
- Aggregate performance metrics consistently mask biome-dependent disparities, as demonstrated by Prithvi-EO-2.0 reaching 90.98% overall macro-F1 on m-EuroSAT but a mean worst-group score of only 83.72%.
Context
Remote sensing foundation models are commonly evaluated using aggregate metrics, which can hide systematic performance disparities across ecological regions. This benchmark addresses that gap by mapping samples into macro-groups and evaluating models under a unified frozen-backbone protocol.
Limitations
Evaluation relies on three random seeds and three random seeds. Code and datasets are available at: https://github.com/aminurhossain/FairRSFM.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 22/10/2026, 05:02:49 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihczdoj6m23js2qvt3re4ok73lrjdnvoclrsxy3xy24l4ecwwnsdm