2imi9/OlmoEarth-v1_2-FT-AWF-Base

This model is fine-tuned from OlmoEarth-v1_2-Base to map land use and land cover in southern Kenya. It follows the recipe and labels of Ai2's OlmoEarth-v1-FT-AWF-Base, with the v1.2 encoder in place of v1. It is a community fine-tune, not an Ai2 release.

Validation accuracy: 90.1% on the 344 validation points of the AWF dataset at the checkpoint kept (epoch 60 of 100), which was chosen on the same points, as Ai2's 89.5% for v1 was; 89.0% at the last epoch.

Training configuration: model.yaml, Ai2's AWF model.yaml with the encoder set to OLMOEARTH_V1_2_BASE and a CSV logger in place of Weights & Biases (rslearn 0.1.17, seed 0). model.ckpt is the checkpoint rslearn model fit wrote; use it with rslearn model predict --config model.yaml --ckpt_path model.ckpt.

This model is licensed under the OlmoEarth Artifact License. It is intended for use in accordance with Ai2's Responsible Use Guidelines.

For more details, please see the documentation on GitHub at https://github.com/allenai/olmoearth_projects/blob/main/docs/awf.md