[arxiv] MFNet

This repo is the official implementation of 'MFNet: Fine-Tuning Segment Anything Model for Multimodal Remote Sensing Semantic Segmentation'.

Usage

You can get the pre-trained model here: https://github.com/facebookresearch/segment-anything?tab=readme-ov-file#model-checkpoints

The core modules are in ./MedSAM/models/ImageEncoder and ./MedSAM/models/sam

The current mode is MMLoRA. You can choose MMAdapter or MMLoRA by the 'mod' hyper-parameter in ./MedSAM/cfg.py, and also need modify the Line 523/524 in SSRS/MFNet /UNetFormer_MMSAM.py and Line 35/36, 42/43 in SSRS/MFNet/train.py. Most of the remaining files come from the original SAM framework and can be ignored.

Run the code by: python train.py

Draw the heatmap by: python test_heatmap.py

Please cite our paper if you find it is useful for your research.

@article{ma2024manet,
  title={MANet: Fine-Tuning Segment Anything Model for Multimodal Remote Sensing Semantic Segmentation},
  author={Ma, Xianping and Zhang, Xiaokang and Pun, Man-On and Huang, Bo},
  journal={arXiv preprint arXiv:2410.11160},
  year={2024}
}