DVCTNet

[MICCAI 2025] Adapting Foundation Model for Dental Caries Detection with Dual-View Co-Training

by Tao Luo *, Han Wu*, Tong Yang, Dinggang Shen, and Zhiming Cui⁺

[Paper] [Project Page]

This repository contains the code and dataset for our paper "Adapting Foundation Model for Dental Caries Detection with Dual-View Co-Training" in MICCAI 2025.

Updates

• [09/2025] Dataset released!
• [09/2025] Source code released!
• [06/2025] Our paper is accepted by MICCAI 2025!

Getting Started

Requirements

• Python 3.9+
• PyTorch 2.1+
• CUDA runtime 11.8 (recommended)
• mmdet 3.3.0
• mmcv 2.1.0

Run the following command to install the required packages with conda:

git clone https://github.com/ShanghaiTech-IMPACT/DVCTNet.git
conda create -f environment.yml

Install all the required packages by pip(unrecommended):

git clone https://github.com/ShanghaiTech-IMPACT/DVCTNet.git
pip install -r requirements.txt

If you have any questions about the installation process, please refer to the mmdetection documentation.

Training and Testing

Before training and testing, the environment variables PYTHONPATH should be appended:

export PYTHONPATH=$PYTHONPATH:./mmdet_custom

To train the model, run the following command:

python tools/train.py \
    configs/models/dvctnet_dinov2_base_fpn_50_epoch.py

or you can use the shell script train.sh in scripts folder to train the model.

bash scripts/train.sh

To test the model, run the following command:

python tools/test.py \
    configs/models/dvctnet_dinov2_base_fpn_50_epoch.py \
    <checkpoint_path> \

To learn more about the configuration files and training and testing process, please refer to the mmdetection documentation.

Dataset

★ Our dataset is available for reserach purpose only. To apply for DVCT dataset, please refer to the dataset website of IMPACT Lab and fill out the form and send the signed e-copy to Tao Luo (email: luotao2024@shanghaitech.edu.cn) and Dr. Zhiming Cui (email: cuizhm@shanghaitech.edu.cn) as well as CC your advisor as mentioned in Sec.5 of the form. We will send you the dataset link and password when recieving the data registration form.

Citation

If you find this code or dataset useful, please cite our paper:

@InProceedings{LuoTao_Adapting_MICCAI2025,
        author = { Luo, Tao and Wu, Han and Yang, Tong and Shen, Dinggang and Cui, Zhiming},
        title = { { Adapting Foundation Model for Dental Caries Detection with Dual-View Co-Training } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15975},
        month = {September},
        page = {44 -- 53}
}

Acknowledgements

• This code is built on mmdetection , ViTAdapter, Timm, and Dinov2.