| Thesis: | Deep Learning-Based Human Body Segmentation on 3D Body Scans |
| Annotation: | Human body part segmentation has an important role in the context of human body analysis. It is often used as an intermediate step in order to solve more complex tasks, which require understanding of human body structure. These days, most of the related research is oriented on machine learning methods, since they proved to outperform the analytical approaches. Performing a body segmentation on 3D input data might be beneficial in comparison to using 2D input images, offering a potential improvement in accuracy by providing the depth information. |
| Goals: | The aim of the thesis is to use machine learning tools to accurately segment a human body into particular body regions, using 3D body scans as an input. As a preliminary step, it is necessary to correctly annotate the real-world 3D data to generate the ground truth for network training. Then, study the machine learning techniques, train and validate selected neural models. Finally, the goal is to evaluate the results and compare the performance to state-of-the-art methods. |