Massively Multi-Person 3D Human Motion Forecasting with Scene Context

TL;DR

This paper introduces SAST, a scene-aware social transformer model that effectively forecasts long-term multi-person 3D human motion by incorporating scene context and interactions, outperforming previous methods on realism and diversity.

Contribution

The paper presents a novel scene-aware social transformer model that models interactions among varying numbers of people and objects for long-term 3D human motion prediction.

Findings

Outperforms previous models in realism and diversity metrics

Effectively models interactions among multiple people and objects

Achieves superior results on the Humans in Kitchens dataset

Abstract

Forecasting long-term 3D human motion is challenging: the stochasticity of human behavior makes it hard to generate realistic human motion from the input sequence alone. Information on the scene environment and the motion of nearby people can greatly aid the generation process. We propose a scene-aware social transformer model (SAST) to forecast long-term (10s) human motion motion. Unlike previous models, our approach can model interactions between both widely varying numbers of people and objects in a scene. We combine a temporal convolutional encoder-decoder architecture with a Transformer-based bottleneck that allows us to efficiently combine motion and scene information. We model the conditional motion distribution using denoising diffusion models. We benchmark our approach on the Humans in Kitchens dataset, which contains 1 to 16 persons and 29 to 50 objects that are visible simultaneously. Our model outperforms other approaches in terms of realism and diversity on different metrics and in a user study. Code is available at https://github.com/felixbmuller/SAST.