The Dataset consists of the source code for defect detection and analysis, and the STEM images, along with the bounding boxes for open loop defects in the images. The source code is written with Matlab and was run with the Matlab r2017a version. Essential toolboxes required to run the code are: Image Processing Toolbox, Statistics and Machine Learning Toolbox, Computer Vision Toolbox. The project has been tested on macOS Sierra, Windows 7, Windows 10 and Ubuntu 17.10 platforms. Some parts of the code require GPU that supports CUDA to function correctly.
The project has three types of files: functions, scripts and data folders. Functions can be divided into several categories, such as data parser, screening method, etc. Scripts can be divided into several categories, such as image augmentation, train cascade object detector.
To run the code, move all the source code to the same folder where all image folders sits. The images folders are located in project folder. Image data folders can be divided into several categories, such as positive images, negative images., eg. "aug_training_positive_images”.
For more information, please refer to the documentation (“README.pdf”) in the dataset.