AI Signal 504
OlmoEarth Studio adds custom embedding exports for Earth observation data
↗
OlmoEarth Studio now lets users compute and export custom embedding vectors from its open-source Earth observation foundation models, delivered as Cloud-Optimized GeoTIFFs with configurable area, time span, encoder variant, resolution, and imagery sources.
This gives engineers working with geospatial data a label-free way to generate compact vector representations for downstream tasks like similarity search and few-shot segmentation. On-demand computation means embeddings reflect specific conditions rather than generic pre-computed archives, and the int8 quantized COG format keeps storage and transfer costs low.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Three encoder variants are available: Nano (128-dim, 1.4M params), Tiny (192-dim, 6.2M params), and Base (768-dim, 89M params), with spatial resolutions from 10m to 80m per pixel.
Embeddings export as Cloud-Optimized GeoTIFFs with int8 quantized vectors and can be dequantized back to floating-point using olmoearth_pretrain.
Embeddings are computed on demand rather than pulled from a pre-computed global archive, allowing monthly embeddings that capture seasonal dynamics.
THE CLUSTER