Home

Awesome

SAM-JOANet

This repo is the implementation of "Joint-Optimized Unsupervised Adversarial Domain Adaptation in Remote Sensing Segmentation with Prompted Foundation Model". We refer to mmsegmentation and mmagic. Many thanks to SenseTime and their two excellent repos.

<table> <tr> <td><img src="PaperFigs\Fig1.png" width = "100%" alt="SAM-JOANet"/></td> </tr> </table>

Dataset Preparation

We select ISPRS (Postsdam/Vaihingen) and CITY-OSM (Paris/Chicago) as benchmark datasets.

We follow ST-DASegNet for detailed dataset preparation.

<table> <tr> <td><img src="PaperFigs\tree_data.png" width = "100%" alt="tree-data"/></td> </tr> </table>

SAM-JOANet

Install

  1. requirements:

    python >= 3.7

    pytorch >= 1.11

    cuda >= 11.7

    This version depends on mmengine and mmcv (2.0.1)

  2. prerequisites: Please refer to MMSegmentation PREREQUISITES.

    cd SAM-JOANet

pip install -e .

chmod 777 ./tools/dist_train.sh

chmod 777 ./tools/dist_test.sh

Training

  1. ISPRS UDA-RSSeg task:

    cd SAM-JOANet

./tools/dist_train.sh ./experiments/SAM_UDA_Sb5PromptSTAdv_bit-b16_upernet.py 2

  2. CITY-OSM UDA_RSSeg task:

    cd SAM-JOANet

./tools/dist_train.sh ./experiments/SAM_UDA_Sb5PromptSTAdv_bit-b16_upernet_P2C.py 2

Testing

Trained with the above commands, you can get your trained model to test the performance of your model.

  1. ISPRS UDA-RSSeg task:

    cd SAM-JOANet

./tools/dist_test.sh ./experiments/SAM_UDA_Sb5PromptSTAdv_bit-b16_upernet.py ./experiments/SAM_UDA_Sb5PromptSTAdv_bit-b16_upernet_results/iter_11000_P2V_66.86.pth

  2. CITY-OSM UDA_RSSeg task:

    cd SAM-JOANet

CUDA_VISIBLE_DEVICES=1 python ./tools/test.py ./experiments/SAM_UDA_Sb5PromptSTAdv_bit-b16_upernet_P2C.py ./experiments/iter_35000_P2C_56.96.pth --show-dir ./P2C_results

The ArXiv version of this paper will be release soon.

If you have any question, please discuss with me by sending email to lyushuchang@buaa.edu.cn.

References

Many thanks to their excellent works