Dual-path network combining CNN and transformer for pavement crack segmentation

Main framework of the proposed network is as follows:

<div align=center> <img src="imgs/network.png" width="400px"> </div>

We have uploaded a model code implemented using PyTorch, which is simple and brief. We also uploaded the example weights trained on the DeepCrack dataset, which can help us better reproduce the model's performance.

Here are two visual examples:

(1) Obtain binary segmentation mask:

<div align=center> <img src="imgs/21_4.png" width="200px"> <img src="results/21_4.jpg" width="200px"> </div>

(2) Obtain the segmentation mask based on the original image:

<div align=center> <img src="imgs/27_2.png" width="200px"> <img src="results/27_2.png" width="200px"> </div>

More details will be described in our paper. If this work is helpful to you, or if you need to use our network in your work, please cite us:

@article{WANG2024105217,
title = {{Dual-path network combining CNN and transformer for pavement crack segmentation}},
journal = {Automation in Construction},
volume = {158},
pages = {105217},
year = {2024},
issn = {0926-5805},
doi = {10.1016/j.autcon.2023.105217},
author = {Jin Wang and Zhigao Zeng and Pradip Kumar Sharma and Osama Alfarraj and Amr Tolba and Jianming Zhang and Lei Wang}
}