• Stepwise Feature Fusion: Local Guides Global
• • packages
  • Dataset
  • Result and Checkpoint
  • • ssformer-S
    • ssformer-L
    • Checkpoints
  • Usage
  • • Test
    • Train
  • Citation

Stepwise Feature Fusion: Local Guides Global

This is the official implementation for Stepwise Feature Fusion: Local Guides Global

packages

• Please see requirements.txt

Dataset

• The dataset we used can be download from here

Result and Checkpoint

ssformer-S

dataset	meanDic	meanIou	wFm	Sm	meanEm	mae
CVC-300	0.887	0.821	0.869	0.929	0.000	0.007
CVC-ClinicDB	0.916	0.873	0.924	0.937	0.000	0.007
Kvasir	0.925	0.878	0.921	0.931	0.000	0.017
CVC-ColonDB	0.772	0.697	0.766	0.844	0.000	0.036
ETIS	0.767	0.698	0.736	0.863	0.000	0.016

ssformer-L

dataset	meanDic	meanIou	wFm	Sm	meanEm	mae
CVC-300	0.895	0.827	0.881	0.933	0.000	0.007
CVC-ClinicDB	0.906	0.855	0.913	0.929	0.000	0.008
Kvasir	0.917	0.864	0.916	0.922	0.000	0.022
CVC-ColonDB	0.802	0.721	0.798	0.860	0.000	0.031
ETIS	0.796	0.720	0.771	0.873	0.000	0.014

Checkpoints

• The checkpoint for ssformer-S can be downloaded from here
• The checkpoint for ssformer-L can be downloaded from here

Usage

Test

1. modified configs/ssformer-S.yaml
   • dataset set to your data path
   • test.checkpoint_save_path : path to your downloaded checkpoint
2. run python test.py configs/ssformer-S.yaml

Train

1. modified configs/train.yaml
   • model.pretrained_path : mit pre-trained checkpoint path
   • other : path to save your training checkpoint and log file
2. run python train.py configs/train.yaml

Citation

@article{wang2022stepwise,
  title={Stepwise Feature Fusion: Local Guides Global},
  author={Wang, Jinfeng and Huang, Qiming and Tang, Feilong and Meng, Jia and Su, Jionglong and Song, Sifan},
  journal={arXiv preprint arXiv:2203.03635},
  year={2022}
}