Bellybutton is a versatile tool that can be used for various purposes, as demonstrated in the example of segmenting a 3D printed photoelastic material in the shape of a granular packing. This material exhibits a birefringence pattern under mechanical stress, making tracking challenging. Bellybutton was trained on images of this system and successfully segmented the particles, with some errors due to image quality issues. The SEG score was used for quantitative analysis, showing good performance. The training data fraction and number of epochs were found to impact the accuracy of the segmentation.
In another example, Bellybutton was used to track the fracturing structure of a lattice of laser-cut acrylic. Despite changes in lighting and focus, Bellybutton was able to accurately track the structure through time. The tool offers options for different types of output, such as innie vs outie or distance-to-edge, which can be useful for different applications.
Overall, Bellybutton proved to be effective in segmenting and tracking complex structures in different materials. The tool’s performance was robust to variations in training data and noise in the images. By adjusting the training data fraction and number of epochs, Bellybutton can provide accurate results even with limited training data. The tool’s versatility and ability to handle various challenges make it a valuable asset for researchers working with complex materials and structures.
Source link
Source link: https://www.nature.com/articles/s41598-024-63906-y