EFSAM

Error-Filterd Segment Anything Model for Few-shot Semantic Segmentation (Visual Computer 2025)

Datasets

• PASCAL-5ⁱ: VOC2012 + SBD

• COCO-20ⁱ: COCO2014

  Download the data lists (.txt files) and put them into the BAM/lists directory.

• Run util/get_mulway_base_data.py to generate base annotations for stage1, or directly use the trained weights.

Getting Started

Our model relies on two models, BAM and SAM.

Firstly, clone BAM, and download datasets and model weights.

Then, clone SAM, and download model weights.

cd BAM-main, make new folder visual, visual/query, visual/output, visual/label, and visual/label2, then make new folder output2pt.

mkdir visual

Copy our test_EF.py, test_EF.sh and test_save.py, test_save.sh to BAM-main, run

sh test_save.sh

Then, cd ../segment-anything-main/notebooks.

Copy our EFSAM.ipynb and EFSAM-multiple.ipynb to ../segment-anything-main/notebooks. Make new file output2.

mkdir output2

Run EFSAM.ipynb or EFSAM-multiple.ipynb to get results and save in ../../BAM-main/output2pt/.

• Note: the mIoU results in EFSAM.ipynb and EFSAM-multiple.ipynb are not the final mIOU.
• Note: Do not use EFSAM-old.ipynb and EFSAM-multiple-old.ipynb.

cd ../../BAM-main, run

sh test_EF.sh

To get the final results by EF-SAM.

Citation

If you find our paper and repo are helpful for your research, please consider citing:

@article{feng2025learning,
  title={Learning few-shot semantic segmentation with error-filtered segment anything model},
  author={Feng, Chen-Bin and Lai, Qi and Liu, Kangdao and Su, Houcheng and Chen, Hao and Luo, Kaixi and Vong, Chi-Man},
  journal={The Visual Computer},
  pages={1--15},
  year={2025},
  publisher={Springer}
}