CrackPropNet is an optical flow-based deep neural network for crack propagation measurement in asphalt concrete cracking tests such as semi-circular bending (SCB) test and indirect tension asphalt cracking test (IDEAL-CT). Simply paint your specimen surface with a random black pattern on top of a thin layer of white paint (tutorial); record a video of the specimen surface during testing (tutorial). Crack propagation can be accurately and efficiently retrived.

Read More (Alternative Link)

git clone https://github.com/zehuiz2/CrackPropNet.git
cd CrackPropNet

# install custom layers
bash install.sh

Currently, the code supports Python 3.8

• numpy
• Pillow
• torch==1.7.1
• torchvision==0.2.2
• tqdm
• glob

CrackPropNet achitecture relies on custom layers Resample2d or Correlation. A pytorch implementation of these layers with cuda kernels are available at ./networks.

The pretrained weights of CrackPropNet is available here.

python inference.py /path/to/input/images --output /path/to/output  --pretrained /path/to/pretrained/model

Sample input images are available at ./img. Reference image should always end with _0.png. Deformed images should end with _*.png. Sample outputs are available at ./output.

If you find this implementation useful in your work, please acknowledge it appropriately and cite the paper:

@article{author = {Zehui Zhu and Imad L. Al-Qadi},
title = {Automated crack propagation measurement on asphalt concrete specimens using an optical flow-based deep neural network},
journal = {International Journal of Pavement Engineering},
volume = {24},
number = {1},
pages = {2186407},
year  = {2023},
publisher = {Taylor & Francis},
doi = {10.1080/10298436.2023.2186407}
}

This implementation is based on the Pytorch implmentation of FlowNetCSS from FlowNet 2.0