Models for generating three-dimensional objects

Subject of this thesis is to explore Generative Adversarial Networks and apply a 3DGAN model to the problem of generating ultrasound images of defects in materials. After an introduction to Deep learning and generative methods, GAN and its derivative 3DGAN models are described. The results of adding another convolutional layer to the 3DGAN structure in order to work with higher resolution 3D models are shown. The model is tested on ModelNet40 dataset of most common 3D objects, and then applied to be used for expanding a dataset of 3D ultrasound images of defects in materials, which in turn is used to train a neural network for detecting defects in production.