Convolutions for Spatial Interaction Modeling
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.
Cite
Text
Su et al. "Convolutions for Spatial Interaction Modeling." Conference on Computer Vision and Pattern Recognition, 2022. doi:10.1109/CVPR52688.2022.00647Markdown
[Su et al. "Convolutions for Spatial Interaction Modeling." Conference on Computer Vision and Pattern Recognition, 2022.](https://mlanthology.org/cvpr/2022/su2022cvpr-convolutions/) doi:10.1109/CVPR52688.2022.00647BibTeX
@inproceedings{su2022cvpr-convolutions,
title = {{Convolutions for Spatial Interaction Modeling}},
author = {Su, Zhaoen and Wang, Chao and Bradley, David and Vallespi-Gonzalez, Carlos and Wellington, Carl and Djuric, Nemanja},
booktitle = {Conference on Computer Vision and Pattern Recognition},
year = {2022},
pages = {6583-6592},
doi = {10.1109/CVPR52688.2022.00647},
url = {https://mlanthology.org/cvpr/2022/su2022cvpr-convolutions/}
}