3D Modelling using Sequential and Convolutional Generative Adversarial Networks

We propose a novel solution for solving a specific problem of generating realistic and varied 3D models for target objects. Existing processes for 3D modelling involve human inspection of CAD models and borrowing parts from them. There have been inspiring advances made by 3D GANs that generate highly varied object shapes but do not adequately attend to objects that are symmetrical or have limited CAD models available as a training data-set. The benefits of the novel model developed by us are three fold: first, it generates realistic shapes by understanding underlying geometry of objects using a limited training data-set; second, it outperforms the 3D-GAN when generating symmetrical 3D object shapes; third, it bridges a research gap by delivering a solution that requires minimal training time and computational resources.