POINT CLOUD IMAGE GENERATION USING GENERATIVE ADVERSARIAL NETWORKS (GAN) AND REINFORCEMENT LEARNING

POINT CLOUD IMAGE GENERATION USING GENERATIVE ADVERSARIAL NETWORKS (GAN) AND REINFORCEMENT LEARNING

B Yashaswi¹, B Jashwanth², S Vamsi Krishna³ V Rishikesh⁴  , R Krishnamoorthy 5

¹Computer Science and Engineering, Bharath Institute of Higher Education and Research, Chennai-73

² Computer Science and Engineering, Bharath Institute of Higher Education and Research, Chennai-73

³ Computer Science and Engineering, Bharath Institute of Higher Education and Research, Chennai-73

⁴ Computer Science and Engineering, Bharath Institute of Higher Education and Research, Chennai-73

5 Assistant Professor, Computer Science and Engineering, Bharath Institute of Higher Education and Research,

Chennai-73

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - Reinforcement learning (RL) agent controls a generative adversarial network in RL-2GAN-Net in a quick and reliable manner. The framework is used for point cloud form generation, which uses the GAN to control 3D meshes into high-fidelity finished shapes. Although a GAN is unstable and difficult to train, the issue is resolved by training the GAN on a latent space representation whose dimension is reduced in comparison to the raw input of a point cloud and using an RL agent to determine the proper input to the GAN to produce the latent space representation of the shape that best fits the current input of a 3D mesh. The point cloud is reliably produced by the recommended pipeline. An RL agent is being trained in this effort to manage the GAN.

Keywords— GAN, Reinforcement Learning, Point Cloud