References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: Convolutions for Spatial Interaction Modeling
(Submitted on 15 Apr 2021 (v1), last revised 8 Jun 2022 (this version, v3))
Abstract: In many different fields interactions between objects play a critical role in determining their behavior. Graph neural networks (GNNs) have emerged as a powerful tool for modeling interactions, although often at the cost of adding considerable complexity and latency. In this paper, we consider the problem of spatial interaction modeling in the context of predicting the motion of actors around autonomous vehicles, and investigate alternatives to GNNs. We revisit 2D convolutions and show that they can demonstrate comparable performance to graph networks in modeling spatial interactions with lower latency, thus providing an effective and efficient alternative in time-critical systems. Moreover, we propose a novel interaction loss to further improve the interaction modeling of the considered methods.
Submission history
From: Zhaoen Su [view email][v1] Thu, 15 Apr 2021 00:41:30 GMT (9611kb,D)
[v2] Wed, 1 Jun 2022 05:14:28 GMT (9857kb,D)
[v3] Wed, 8 Jun 2022 15:28:54 GMT (10049kb,D)
Link back to: arXiv, form interface, contact.