Pose prediction is to predict future poses given a window of previous poses. In this paper, we propose a new problem that predicts poses using 3D positions of skeletal sequences.Different from the traditional pose prediction based on mocap frames, this problem is convenient to use in real applications due to its simple sensors to capture data. We also present a new framework, pseudo-image sequence evolution-based 3D pose prediction, to address this new problem. Specifically, a skeletal representation is proposed by transforming a 3D skeletal sequence into an image sequence, which can model different correlations among different joints. With this image-based skeletal representation, we model the pose prediction as the evolution of an image sequence. Moreover, a novel inference network is proposed to predict multiple future poses in a non-recursive manner using decoders with independent parameters. In contrast to the recursive sequence-to-sequence model, we can improve the computational efficiency and avoid error accumulations significantly. Extensive experiments are carried out on two benchmark datasets (e.g., G3D and FNTU). The proposed method achieves state-of-the-art performance on both datasets, which demonstrates the effectiveness of our proposed method.

姿势预测是在给定先前姿势的窗口的情况下预测未来姿势。在本文中，我们提出了一个新的问题，该问题使用骨骼序列的3D位置预测姿势。与基于mocap帧的传统姿势预测不同，该问题由于其简单的传感器来捕获数据而在实际应用中使用起来很方便。我们还提出了一种新框架，即基于伪图像序列演化的3D姿态预测，以解决这一新问题。具体而言，通过将3D骨骼序列转换为图像序列来提出骨骼表示，该图像序列可以对不同关节之间的不同相关性进行建模。使用这种基于图像的骨骼表示，我们将姿势预测建模为图像序列的演化。而且，提出了一种新颖的推理网络，以使用具有独立参数的解码器以非递归方式预测多个未来姿态。与递归序列到序列模型相比，我们可以提高计算效率并避免错误累积。在两个基准数据集（例如G3D和FNTU）上进行了广泛的实验。所提出的方法在两个数据集上均达到了最先进的性能，这证明了所提出方法的有效性。