Model weights for SAR-W-SimMIM
Model weigits for SimMIM pretrained with ALOS-2 data with SAR weighted loss, and finetuned model for Segmentation with JAXA-LULC labels
By Nevrez Imamoglu*, Ali Caglayan*, Toru Kouyama.
The repository includes both model weights:
- Pretrained model (encoder) with ALOS-2 SAR data with SimMIM Self-supervised pretraining way
- Finetuned model for downstream segmentation using JAXA-LULC
https://github.com/gsvrg/SAR-W-SimMIM
*Nevrez Imamoglu and Ali Caglayan has contributed to this work (both research and repository) equally as co-first authors.
Acknowledgements
This work supported by AIST policy-based budget project “R&D on Generative AI Foundation Models for the Physical Domain".The ALOS-2 original data are copy-righted by JAXA and provided under the JAXA-AIST agreement. We used ABCI 3.0 provided by AIST and AIST Solutions with support from “ABCI 3.0 Development Acceleration Use”.
If you use this model and publish any work, presentation, or other thing, please state that source data for the foundation models are ALOS-2 and the ALOS-2 original data are copy-righted by JAXA and provided by JAXA under the JAXA-AIST agreement.