Multi-Modal Foundation · allenai
OlmoEarth v1.1 Tiny
A 12M-parameter foundation model for remote-sensing tasks across Sentinel-1, Sentinel-2, and Landsat imagery.
What OlmoEarth v1.1 Tiny is
OlmoEarth v1.1 Tiny is a compact Earth-observation foundation model developed by allenai. Designed for applications across earth science, climate research, and conservation, it supports remote-sensing analysis on both static imagery and multi-temporal sequences.
The model serves as a lightweight variant engineered for large-scale data processing and on-device satellite-imagery inference. It processes imagery from Sentinel-1, Sentinel-2, and Landsat platforms, providing general-purpose representations that can be deployed across various downstream environmental workflows.
How it works
OlmoEarth v1.1 Tiny is built upon a ViT-Tiny (Vision Transformer Tiny) architecture containing 12 million parameters. It ingests optical and radar data from Sentinel-1, Sentinel-2, and Landsat sensors, accepting individual images or multi-temporal image series to generate dense visual representations for downstream transfer learning.
Key facts
- Architecture
- ViT-Tiny
- Parameter count
- 12M
- Supported sensors
- Sentinel-1, Sentinel-2, and Landsat
- Input formats
- Single images and image time series
What you can use it for
- Satellite Image Segmentation
- Fine-tuning the vision transformer on remote-sensing scenes to perform pixel-level semantic classification.
- Remote Sensing Embedding Extraction
- Generating compact numerical feature representations from satellite captures or image series for downstream analysis.
- Time-Series Environmental Monitoring
- Processing temporal series from Sentinel or Landsat instruments to assess environmental, climatic, or surface changes over time.
- On-Device and High-Throughput Inference
- Executing inference pipelines in resource-constrained environments or across large volumes of satellite imagery.
Run it here
We list this model but cannot run it yet: no pipeline_tag — needs a human.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Contains 12 million parameters, allowing deployment in on-device and high-throughput settings.
- Operates natively across multiple Earth-observation constellations, including Sentinel-1 radar as well as Sentinel-2 and Landsat optical imagery.
- Accepts both single-date satellite acquisitions and multi-image time series as inputs.
Watch out for
- Requires fine-tuning or adaptation pipelines to perform specific downstream tasks such as semantic segmentation.
- Input capability is restricted to data modalities aligned with Sentinel-1, Sentinel-2, and Landsat sensors.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use OlmoEarth v1.1 Tiny commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
What satellite data types does OlmoEarth v1.1 Tiny accept?
The model accepts radar imagery from Sentinel-1, optical imagery from Sentinel-2, and optical imagery from Landsat, supporting both individual scenes and image time series.
What architecture and size does this model use?
It is based on a ViT-Tiny architecture comprising 12 million parameters.
Can this model be fine-tuned for downstream mapping tasks?
Yes, it can be fine-tuned for downstream remote-sensing applications such as segmentation using packages including rslearn and olmoearth_run.
Can the model output embeddings directly?
Yes, the model can be used to extract remote-sensing embeddings directly from compatible input imagery.
Written from OlmoEarth v1.1 Tiny's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.