SAUGE

Publication

SAUGE: Taming SAM for Uncertainty-Aligned Multi-Granularity Edge Detection (AAAI 2025, Arxiv)

Xing Liufu, Chaolei Tan, Xiaotong Lin, Yonggang Qi, Jinxuan Li, Jian-Fang Hu

Environmental Setup

Please follow the official installation steps for SAM: https://github.com/facebookresearch/segment-anything.

Dataset

Please follow the guide of UAED for preparing BSDS500 and Multicue datasets.

Inference

Run the following command: python demo.py

We have provided the checkpoint for the basic version of SAUGE (trained on the BSDS500 dataset). Please adjust the dataset path and the checkpoint path for SAM in demo.py accordingly.

Training

Please modify the relevant data/hyperparameters and other configurations in train.py, and then run the following command for training: nohup python -m torch.distributed.launch --nproc_per_node=4 --master_port 123 train.py > nohup.log 2>&1 &

Pre-computed Results and Evaluation

We have organized the precomputed results for BSDS500 datasets under various settings in the eval_res directory. To reproduce the results reported in the paper, modify the relevant paths and configurations in eval_res/best_ods_ois.py and then run python eval_res/best_ods_ois.py.

Acknowledgement & Citing SAUGE

The work is highly based on the Segment Anything and UAED. We gratefully acknowledge their excellent work.

If this project supports your research, please consider including a citation in your publications.

Liufu, X., Tan, C., Lin, X., Qi, Y., Li, J., & Hu, J.-F. (2025). SAUGE: Taming SAM for Uncertainty-Aligned Multi-Granularity Edge Detection. Proceedings of the AAAI Conference on Artificial Intelligence, 39(6), 5766-5774. https://doi.org/10.1609/aaai.v39i6.32615