This project provides the code and results for 'Ordered Cross-Scale Interaction Network for No-Service Rail Surface Defect Segmentation', IEEE TIM, 2025. IEEE Homepage

python 3.8 + pytorch 1.9.0

We provide segmentation maps (code: gaw6) of our OCINet and 19 compared methods on the NRSD-MN dataset.

Download pvt_v2_b2.pth (code: sxiq), and put it in './model/'.

Download dataset.zip (code: ajsz), and unzip it. Then, we use data_aug.m for data augmentation.

Modify paths of datasets, then run train_OCINet.py.

Note: Our main model is under './model/OCINet_models.py'

1. Download the pre-trained model (code: hryz) on the NRSD-MN dataset, and put it in './models/'.

2. Modify paths of pre-trained models and datasets.

3. Run test_OCINet.py.

You can use 'evaluation_Dice.py' to evaluate the segmentation maps.

2025_IEEE SJ_DiffRSDI

2023_IEEE TIM_NaDiNet

2022_IEEE TIM_TSERNet

2022_Measurement_CSEPNet

    @ARTICLE{Li_2025_OCINet,
            author = {Gongyang Li and Xiaofei Zhou and Hongyun Li},
            title = {Ordered Cross-Scale Interaction Network for No-Service Rail Surface Defect Segmentation},
            journal = {IEEE Transactions on Instrumentation and Measurement},
            volume = {74},
            pages = {5033210},
            year = {2025},
            month = {Jun.},
            }

If you encounter any problems with the code, want to report bugs, etc.

Please get in touch with me at lllmiemie@163.com or ligongyang@shu.edu.cn.