Reproduce “Towards Consistent Geospatial Representations: Aligning One Encoder to Many Foundation Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Arthur Ouaknine, David Rolnick

GeoMixture trains a ViT student separately on each dataset to align spatial tokens with frozen representations from five geospatial foundation models. The student is evaluated as a frozen encoder. Across nine PANGAEA datasets, GeoMixture-SUM ties TerraMind-L at average rank 3.89, but TerraMind-L has higher average mIoU: 57.05 versus 56.49. This supports cross-task consistency, not universal dominance or transfer across datasets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.