SD-AANet

The code is for the paper "A Self-Distillation Embedded Supervised Affinity Attention Model for Few-Shot Segmentation" [arxiv]

Overview

• config/ includes config files
• lists/ includes train/validation list files
• model/ includes related model and module
• util/ includes data processing, seed initialization

Usage

Requirements

python==3.7, torch==1.8, scipy, opencv-python, tensorboardX

Dataset

Please prepare related datasets:

• Pascal-5ⁱ (VOC 2012, SBD)
• COCO-20ⁱ (COCO 2014)

Specify the paths of datasets in config files, including data root and list files paths

Pre-trained models

We provide 8 pre-trained models: 4 ResNet-50 based models for Pascal-5i and 4 ResNet-101 based models for COCO-20i

• Download the pre-trained models [Pre-trained models]
• Specify the split setting, shot setting and path of weights in config files

Test and Train

• Use the following command for testing

  sh test.sh {data} {split_backbone}
  

E.g. Test SD-AANet with ResNet50 on the split 0 of PASCAL-5i:

sh test.sh pascal split0_resnet50

• Use the following command for training

  sh train.sh {data} {split_backbone}
  

E.g. Train SD-AANet with ResNet50 on the split 0 of PASCAL-5i:

sh train.sh pascal split0_resnet50

Citation

If you have any question, please discuss with me by sending email to [email protected]

Please consider citing the paper if you find it useful:

@article{DBLP:journals/corr/abs-2108-06600,
  author    = {Qi Zhao and
               Binghao Liu and
               Shuchang Lyu and
               Xu Wang and
               Yifan Yang},
  title     = {A Self-Distillation Embedded Supervised Affinity Attention Model for
               Few-Shot Segmentation},
  journal   = {CoRR},
  url       = {https://arxiv.org/abs/2108.06600},
}

References

The code is based on PFENet and kd-pytorch. Thanks for their great work!