Skip to content

Latest commit

 

History

History
30 lines (24 loc) · 1.47 KB

README.md

File metadata and controls

30 lines (24 loc) · 1.47 KB

SAMWS

This is the repository for "A Segment Anything Model based weakly supervised learning method for crop mapping using Sentinel-2 time series images".

Finetuning with adapters

Download SAM checkpoint, we used ViT-B SAM model.

The main finetuning code is in adapter_finetune.py. Some example data is provided in the repository.

Generating pseudo labels

The codes of generating pseudo labels by different weak annotations lie in pseudoLabels_generate_XXX.py. Our image classification model--attention-based U-Net is in cls_model directory.

Training a segmentation model

The segmentation model this study uses is U-TAE.

Citation

@article{sun2024segment,
  title={A Segment Anything Model based weakly supervised learning method for crop mapping using Sentinel-2 time series images},
  author={Sun, Jialin and Yan, Shuai and Yao, Xiaochuang and Gao, Bingbo and Yang, Jianyu},
  journal={International Journal of Applied Earth Observation and Geoinformation},
  volume={133},
  pages={104085},
  year={2024},
  publisher={Elsevier}
}

Acknowledgement

Spectial thanks to scholars from Meta and Junde Wu, as well as his fellows, for their implementation of adapters in SAM.