Diverse Human Motion Prediction Guided by Multi-level Spatial-Temporal Anchors
Diverse Human Motion Prediction Guided by Multi-level Spatial-Temporal Anchors
复制标题
DOI：
10.1007/978-3-031-20047-2_15
复制
发表时间：
2023-02
期刊：
影响因子：
--
通讯作者：
Sirui Xu;Yu-Xiong Wang;Liangyan Gui
中科院分区：
文献类型：
--
作者：
Sirui Xu;Yu-Xiong Wang;Liangyan Gui
在给定一系列历史姿势的情况下预测不同的人体运动已经受到越来越多的关注。尽管进展迅速，现有的工作捕捉多模态性质的人体运动主要是通过基于似然性的采样，其中模式崩溃已被广泛观察到。在本文中，我们提出了一个简单而有效的方法，解开随机抽样的代码与一个确定性的可学习的组件命名锚，以提高样本精度和多样性。该方法将视差因子分解为空间锚和时间锚，对时空视差进行了可解释的控制。原则上，我们的时空锚点采样（STARS）可以应用于不同的运动预测。在这里，我们提出了一种交互增强的时空图卷积网络（IE-STGCN），它对人类运动的先验知识进行编码（例如，大量的实验表明，我们的方法在随机性和确定性预测方面都优于现有技术，这表明它是建模人体运动的统一框架。我们的代码和预训练模型可以在https://github.com/Sirui-Xu/STARS上找到。
Predicting diverse human motions given a sequence of historical poses has received increasing attention. Despite rapid progress, existing work captures the multi-modal nature of human motions primarily through likelihood-based sampling, where the mode collapse has been widely observed. In this paper, we propose a simple yet effective approach that disentangles randomly sampled codes with adeterministic learnable component named anchorsto promote sample precision and diversity. Anchors are further factorized into spatial anchors and temporal anchors, which provide attractivelyinterpretablecontrol over spatial-temporal disparity. In principle, our spatial-temporal anchor-based sampling (STARS) can be applied to different motion predictors. Here we propose an interaction-enhanced spatial-temporal graph convolutional network (IE-STGCN) that encodes prior knowledge of human motions (e.g., spatial locality), and incorporate the anchors into it. Extensive experiments demonstrate that our approach outperforms state of the art in both stochastic and deterministic prediction, suggesting it as aunifiedframework for modeling human motions. Our code and pretrained models are available at https://github.com/Sirui-Xu/STARS.