shiweiluo99/tunnel-crack-segmentation-dataset

Complex Tunnel Crack Dataset (CTCD)

292 tunnel lining images (256 × 256) with pixel-level crack masks, collected from 52 highway tunnels in Yunnan and Zhejiang, China, using a high-speed tunnel inspection vehicle. Crack pixels were manually annotated in Photoshop with a 1–2 px brush sized by crack width.

This dataset was introduced in the CraSAM paper (Structural Control and Health Monitoring, 2025), where it is used to evaluate cross-scenario generalization of foundation-model-based crack segmentation.

📄 Paper: https://doi.org/10.1155/stc/6269747

Usage

from datasets import load_dataset
ds = load_dataset("shiweiluo99/tunnel-crack-segmentation-dataset")

License

CC-BY-4.0. Images and annotations by the authors.

Citation

@article{luo2025crasam,
  title   = {Multiscenario Generalization Crack Detection Network Based on the Visual Foundation Model},
  author  = {Luo, Shiwei and Xie, Xiongyao and Zhou, Biao and Zeng, Kun and Guo, Jun},
  journal = {Structural Control and Health Monitoring},
  volume  = {2025},
  number  = {1},
  pages   = {6269747},
  year    = {2025},
  doi     = {10.1155/stc/6269747}
}