FlowMoP generates stochastic multi-person motion prediction
Original title: FlowMoP: Stochastic Multi-Person Motion Prediction
FlowMoP uses flow matching for diverse multi-agent motions achieving SOTA on CMU-Mocap and 3DPW.
Multi-person motion prediction is stochastic with many plausible continuations. FlowMoP jointly generates diverse global trajectories and poses via conditional flow matching. B-spline reparameterization compresses state space enforcing smoothness. Dual-stream transformer conditions on neighbor motion for interaction-aware forecasting. It outperforms deterministic baselines on error metrics and establishes a baseline on WorldPose soccer data.
- SOTA on CMU-Mocap and 3DPW
- Full predictive distribution unlike deterministic
- First baseline on WorldPose soccer
- Social conditioning along interaction directions
- Dual-stream transformer for neighbor conditioning
From the original article
1 min read at arxiv.orgPredicting the future motion of multiple people is inherently stochastic: the same observed history admits many plausible continuations depending on intent, coordination, and social context. Yet the dominant paradigm in multi-person 3D motion prediction remains deterministic, with existing methods producing a fixed number of outputs per agent.
By Aadya Agrawal · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.