Upsampling 4D Point Clouds of Human Body via Adversarial Generation

Time varying sequences of 3D point clouds, or 4D point clouds, are acquired at an increasing pace in several applications (e.g., LiDAR in autonomous or assisted driving). In many cases, such volume of data is transmitted, thus requiring that proper compression tools are applied to either reduce the resolution or the bandwidth. In this paper, we propose a new solution for upscaling of time-varying 3D video point clouds. Our model consists of a specifically designed Graph Convolutional Network that combines Dynamic Edge Convolution and Graph Attention Networks for feature aggregation in a Generative Adversarial setting. To make these modules work in synergy, we present a specific way to sample dense point clouds and provide each node with enough features of its neighbourhood to generate new vertices. Compared to other solutions in the literature that address the same task, our proposed model is capable of obtaining similar results in terms of quality of reconstruction, while using a substantially lower number of parameters ( \(\simeq \) 300KB).