Twin Models for High Resolution Visual Inspections

IC-SHM 2021, Project 2

A team effort by PhD candidates:

Kareem Eltouny and Seyedomid Sajedi

Jan 2022, University at Buffalo (SUNY), Department of Civil, Structural and Environmental Engineering

Introduction

This is the official repository for the code and models used by UB-SHM team in IC-SHM 2021, project 2. Models are evalauted on the QuakeCity benchmark dataset. You can find further details about the twin models in this report.

Models and code

The code and models are developed using . Trained models and weights can be found from releases.

Name	Segmentation Task
	Component type
	Component damage severity
	Cracks, rebar exposure, spalling
	Cracks, rebar exposure, spalling
	Cracks, rebar exposure, spalling

TRS-Net

DmgFormer

Acknowledgements

• The Swin Transformer backbone implementation is a slight modification of the official repository from Microsoft
• TRS-Net is based on ResNeSt[1] and U-Net++[2] implementations from SMP and timm-models

References

[1] H. Zhang et al., "Resnest: Split-attention networks," arXiv preprint arXiv:2004.08955, 2020.

[2] Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh, and J. Liang, "Unet++: A nested u-net architecture for medical image segmentation," in Deep learning in medical image analysis and multimodal learning for clinical decision support: Springer, 2018, pp. 3-11.

[3] Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S. and Guo, B. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 10012-10022.