Dataset for TOPO-Loss for continuity-preserving crack detection using deep learning
This is the dataset related to the crack detection algorithm proposed by Pantoja-Rosero et, al (2022) in the article "TOPO-Loss for continuity-preserving crack detection using deep learning"
The dataset is formed by 3 main folders described as follows:
- data: contains the WILD image data set used to train the different models. In this folder training and validation sets are formed by images, anotation as binary image and its skeleton (used to compute distance maps). Further a test set is given formed by full sized images. Labels used to compute CPP metric with image patches from test set are also provided.
- models: contains the trained models using the WILD dataset and four combination of losses --- MSE, TOPO, DICE+TOPO and MSE+TOPO (best results).input pointcloud with normals necessary to run polyfit pipeline to obtain LOD2 models. Used parameters are given as well as the outputs. This for each example presented in the article.
- results: contains the output of the inference using the full sized test images and the MSE+TOPO model presented in the article.