Balanced Feature Fusion for Grouped 3D Pose Estimation

3D human pose estimation by grouping body joints according to anatomical relationship is currently a popular and effective method. For grouped estimation, fusing features of different groups together effectively the key step ensure integrity whole prediction. However, existing methods for feature fusion between require large number network parameters, thus are often computational expensive. In this paper, we propose simple yet efficient method that can improve accuracy esti- mation while fewer parameters less calculations. Experiments have shown our proposed outperforms previous state-of-the-art results on Human3.6M dataset.