Learning to predict diverse human motions from a single image via mixture density networks

作者：

Highlights：

• We propose a novel perspective for predicting human motion from the weakly informed conditions (i.e., a single image) by giving diverse plausible predictions.

• We model human motion from the image domain to 3D space under mixture density networks and formulate energy-based priors to improve the prediction performance.

• We quantitatively and qualitatively report extensive experimental results on Human3.6M and MPII datasets, and show the robustness of our method in terms of prediction accuracy and diversity.

摘要

•We propose a novel perspective for predicting human motion from the weakly informed conditions (i.e., a single image) by giving diverse plausible predictions.•We model human motion from the image domain to 3D space under mixture density networks and formulate energy-based priors to improve the prediction performance.•We quantitatively and qualitatively report extensive experimental results on Human3.6M and MPII datasets, and show the robustness of our method in terms of prediction accuracy and diversity.

论文关键词：Human motion prediction,Mixture density networks,Energy-based prior

论文评审过程：Received 22 December 2021, Revised 21 July 2022, Accepted 22 July 2022, Available online 28 July 2022, Version of Record 6 August 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109549