Diverse Human Motion Prediction Guided by Multi-Level Spatial-Temporal Anchors

TL;DR

This paper introduces a novel spatial-temporal anchor-based sampling method for human motion prediction, improving diversity and accuracy over likelihood-based approaches by disentangling sampled codes with interpretable anchors.

Contribution

It proposes a unified framework using spatial-temporal anchors and an interaction-enhanced graph network to enhance human motion prediction, addressing mode collapse issues.

Findings

Outperforms state-of-the-art in stochastic and deterministic predictions

Provides interpretable control over spatial-temporal disparity

Demonstrates effectiveness across various motion prediction tasks

Abstract

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 a deterministic learnable component named anchors to promote sample precision and diversity. Anchors are further factorized into spatial anchors and temporal anchors, which provide attractively interpretable control 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 a unified framework for modeling human motions. Our code and pretrained models are available at https://github.com/Sirui-Xu/STARS.