GARNet: Graph Attention Residual Networks Based on Adversarial Learning for 3D Human Pose Estimation

Recent studies have shown that, with the help of complex network architecture, great progress has been made in estimating the pose and shape of a 3D human from a single image. However, existing methods fail to produce accurate and natural results for different environments. In this paper, we proposed a novel adversarial learning approach and studied the problem of learning graph attention network for regression. Graph Attention Residual Networks (GARNet), which processes regression tasks with graphic-structured data, learns to capture semantic information, such as local and global node relationships, through end-to-end training without additional supervision. The adversarial learning module is implemented by a novel multi-source discriminator network to learn the mapping from 2D pose distribution to 3D pose distribution. We conducted a comprehensive study to verify the effectiveness of our method. Experiments show that the performance of our method is superior to that of most existing techniques.