3DHumanGAN: 3D-Aware Human Image Generation with 3D Pose Mapping

Human image generation is a long-standing topic in computer We present 3DHumanGAN, a 3D-aware generative adversarial vision and graphics with applications across multiple network that synthesizes photo-like images of fullbody areas of interest including movie production, social networking humans with consistent appearances under different and e-commerce. Compared to physically-based methods, view-angles and body-poses. To tackle the representational data-driven approaches are preferred due to the photolikeness and computational challenges in synthesizing the articulated of their results, versatility and ease of use [60]. In structure of human bodies, we propose a novel generator this work, we are interested in synthesizing full-body human architecture in which a 2D convolutional backbone is modulated images with a 3D-aware generative adversarial network by a 3D pose mapping network. The 3D pose mapping (GAN) that produces appearance-consistent images under network is formulated as a renderable implicit function conditioned different view-angles and body-poses.