Home

Awesome

MuGE

We add code for MuGE.
MuGE: Multiple Granularity Edge Detection pdf
Caixia Zhou, Yaping Huang, Mengyang Pu, Qingji Guan, Ruoxi Deng and Haibin Ling
CVPR2024

UAED

The Treasure Beneath Multiple Annotations: An Uncertainty-aware Edge Detector
Caixia Zhou, Yaping Huang, Mengyang Pu, Qingji Guan, Li Huang and Haibin Ling
CVPR 2023

Preparing Data

The processed dataset is from LPCB, you can download the used matlab code and processed data from the Baidu disk, the code is 3tii. The complete processed BSDS training dataset can be downloaded from the Google disk. Training data for the Multicue dataset can be downloaded from the Quark Disk.

Checkpoint

BSDS with single scale for UAED: Quark disk or Google disk
VOC pretrain model for UAED: Quark disk or Google disk Pretrain granularity network for MuGE: Google disk BSDS with scale for MuGE: Google disk

Results

UAED Results for BSDS under a single-scale setting can be found here.

Start

UAED:

python train_uaed.py

MuGE: first download the checkpoint, then revise line 39 as the checkpoint path.

python train_muge.py

Best ODS and OIS evaluation for MuGE

  1. Run python test_muge.py to obtain the results under different granularities.
  2. Test ODS and OIS for each granularity as normal.
  3. Run eval_muge_best/best_ods_ois.py to obtain the ODS and OIS value.
  4. Run eval_muge_best/select_best_ois_png.py to obtain the selected pictures for best OIS.
  5. Select the threshold for best ODS from the best_ods_0.1/nms-eval/eval_bdry_thr.txt, revise line 8 in eval_muge_best/select_best_ods_png.py and run eval_muge_best/select_best_ods_png.py to select the best pictures for best ODS.

Acknowledgement & Citation

The dataset is highly based on the LPCB, and the code is highly based on RCF_Pytorch_Updated and segmentation_models.pytorch. Many thanks for their great work.
Please consider citing this project in your publications if it helps your research.

@inproceedings{zhou2023treasure,
  title={The treasure beneath multiple annotations: An uncertainty-aware edge detector},
  author={Zhou, Caixia and Huang, Yaping and Pu, Mengyang and Guan, Qingji and Huang, Li and Ling, Haibin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={15507--15517},
  year={2023}
}

@inproceedings{zhou2024muge,
  title={MuGE: Multiple Granularity Edge Detection},
  author={Zhou, Caixia and Huang, Yaping and Pu, Mengyang and Guan, Qingji and Deng, Ruoxi and Ling, Haibin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2024}
}

@inproceedings{deng2018learning,
  title={Learning to predict crisp boundaries},
  author={Deng, Ruoxi and Shen, Chunhua and Liu, Shengjun and Wang, Huibing and Liu, Xinru},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={562--578},
  year={2018}
}