Dear audience of Stackoverflow,
I am a student and I have a task to segment concrete components (coarse aggregates, voids and cracks) from X-ray CT images. I have started to study the DL and I think that the best way in my segmentation case is using CNN models such as Mask R-CNN and U-net. I found the articles that relate to my research task ("A two-phase approach using Mask R-CNN and 3D U-net for High-Accuracy Automatic Segmentation of Pancreas in CT Imaging" and "Meso-structure segmentation of concrete CT images based on mask and regional convolutional neural network" and I plan to base my research on them. I have a question regarding training data preparation. Before, I manually segmented all concrete components from the longitudinal cross-section CT image of the concrete cylinder (side view of the cylinder). The image size varies from 200x400 to 400x700 pixels. But I want to train the model that it can segment these concrete particles from the cross-section of CT images (up view on the cylinder). These images are square images of 512x512 size, where the diameter of the cylinder diameter size is presented. Thus, my question is: can I use the longitudinal images for model training that will segment the target objects from cross-section (square) images subsequently? Thank you in advance.
It is my first time trying DL model segmentation for CT images. Therefore, I would like to get advice on how to properly prepare the training data set.
