Dyadic Human Motion Prediction

Abstract

Prior work on human motion forecasting has mostly focused on predicting the future motion of single subjects in isolation from their past pose sequence. In the presence of closely interacting people, however, this strategy fails to account for the dependencies between the different subject's motions. In this paper, we therefore introduce a motion prediction framework that explicitly reasons about the interactions of two observed subjects. Specifically, we achieve this by introducing a pairwise attention mechanism that models the mutual dependencies in the motion history of the two subjects. This allows us to preserve the long-term motion dynamics in a more realistic way and more robustly predict unusual and fast-paced movements, such as the ones occurring in a dance scenario. To evaluate this, and because no existing motion prediction datasets depict two closelyinteracting subjects, we introduce the LindyHop600K dance dataset. Our results evidence that our approach outperforms the state-of-the-art single person motion prediction techniques.

Reconstructing Close Human Interactions from Multiple Views

“…We only utilize temporal input during the inference stage. Future work could incorporate temporal motion prior learning [Rempe et al 2021] and spatial motion prior learning [Adeli et al 2020;Guo et al 2022;Katircioglu et al 2021]. Such enhancement could potentially allow the model to gain more understanding of motion patterns and improve the overall accuracy of the pose estimation.…”