This repo is the official implementation for:
- Multi Kernel Positional Embedding ConvNeXt for Polyp Segmentation (RIVF 2022).
- PEFNet: Positional Embedding Feature for Polyp Segmentation (MMM 2023).
Detail of each model modules can be found in original paper. Please citation if you use our implementation for research purpose.
Architecutre of PEFNet and PEFNet with Multi-Kernel module:
Our implementation is on Python 3.9 , please make sure to config your environment compatible with the requirements.
To install all packages, use requirements.txt file to install. Install with pip by the following command:
pip install -r requirements.txt
All packages will be automatically installed.
All of configs for training and benchmark are in ./config/ folder. Please take a look for tuning phase.
For training, use train.py file for start training.
The following command should be used:
python train.py
For benchmar, use test.py file for start testing.
The following command should be used:
python test.py
Note: you should fix model_path for your model path and directory to your benchmark dataset.
The weight will be update later.
You can use Kvasir-SEG dataset for training, or CVC-clinic DB for training.
The IOU score on SOTA for Kvasir-SEG, this is our best model:
| Model | IOU | Dice Coef |
|---|---|---|
| PEFNet (MMM 2023) | 82.01 | 88.02 |
| PEFNet + Multi-Kernel (RIVF 2022) | 81.63 | 88.18 |
Some results of visualization:
@inproceedings{nguyen2022multi,
title={Multi Kernel Positional Embedding ConvNeXt for Polyp Segmentation},
author={Nguyen-Mau, Trong-Hieu and Trinh, Quoc-Huy and Bui, Nhat-Tan and Tran, Minh-Triet and Nguyen, Hai-Dang},
booktitle={2022 RIVF International Conference on Computing and Communication Technologies (RIVF)},
pages={731--736},
year={2022},
organization={IEEE}
}
@inproceedings{10.1007/978-3-031-27818-1_20,
title={PEFNet: Positional Embedding Feature for Polyp Segmentation},
author={Nguyen-Mau, Trong-Hieu and Trinh, Quoc-Huy and Bui, Nhat-Tan and Thi, Phuoc-Thao Vo and Nguyen, Minh-Van and Cao, Xuan-Nam and Tran, Minh-Triet and Nguyen, Hai-Dang},
booktitle={MultiMedia Modeling},
pages={240--251},
year={2023},
publisher={Springer Nature Switzerland}
}