Data for: Computer Vision-based Concrete Crack Detection using U-Net Fully Convolutional Networks

Published: 24 April 2019| Version 1 | DOI: 10.17632/c7cpnw32j6.1
Contributors:
Zhenqing Liu, Guowei Qian, Wei Wang

Description

The three subdirectories under the ./code/data/ directory respectively store the train set data, the verification set data, and the test set data. The ./code/cnn_with_slide_window/ directory stores the code for Cha’s CNN. cleandata.ipynb preprocesses the data and stores it in the ./data/ directory. The ./logs/ and ./trained_models/ directory store the training process log and the trained model. dataset.py is the I/O related code. Myutils.py is the tool class code. model.py defines the network structure. train.py is the training code and predict.py is the prediction code. The ./code/cunet/ directory stores U-Net implementation code. Dice_coeff_loos.py, focalloss.py and lovasz_losses.py are the user defined loss functions. The functions of other files in ./code/cunet/ are similar to the of files with the same name in ./code/cnn_with_slide_window/. Environment is Python 3.6 and Pytorch 1.0 (> 0.4). Display memory requirement is 5000MB. Recommended minimum hardware is GTX1060 6GB.

Files

Categories

Concrete Cracking

Licence