PROJECT TITLE :

Generative VoxelNet: Learning Energy-Based Models for 3D Shape Synthesis and Analysis

ABSTRACT:

Understanding the 3D physical world requires having access to 3D data that has a rich amount of information regarding the geometry of objects and scenes. In light of the recent proliferation of large-scale 3D datasets, it is becoming increasingly important to possess a potent 3D generative model for the purpose of conducting 3D shape analysis and synthesis. A deep three-dimensional energy-based model is proposed as a means of representing volumetric shapes in this paper. An "analysis by synthesis" approach is taken during the training of the model to achieve the maximum possible likelihood. The conditional model can be applied to 3D object recovery and super resolution; the model can serve as a building block in a multi-grid modeling and sampling framework for high resolution 3D shape synthesis; and the model can be used. The six benefits of the proposed model are as follows: first, unlike GANs and VAEs, the model training does not rely on any auxiliary models; second, the model can synthesize realistic 3D shapes by Markov chain Monte Carlo (MCMC); third, the The results of the experiments show that the proposed model is capable of generating high-quality 3D shape patterns and has the potential to be useful for various types of 3D shape analysis.