Assessing the Data-Centric Robustness of a Crack Tip Detection Model through Artificial Data Pollution¶

This repository provides the experimental framework used in the study Assessing the Data-Centric Robustness of a Crack Tip Detection Model through Artificial Data Pollution.

The objective of the work is to evaluate the robustness of crack-tip segmentation methods under controlled data degradation scenarios. The CrackMNIST dataset is subjected to parameterized perturbations (noise, blur, and occlusions), and the performance of a U-Net segmentation model is assessed under varying degrees of input quality.

Please refer to the README for installation and usage instructions.