English

Exploiting map information for self-supervised learning in motion forecasting

Computer Vision and Pattern Recognition 2022-10-11 v1

Abstract

Inspired by recent developments regarding the application of self-supervised learning (SSL), we devise an auxiliary task for trajectory prediction that takes advantage of map-only information such as graph connectivity with the intent of improving map comprehension and generalization. We apply this auxiliary task through two frameworks - multitasking and pretraining. In either framework we observe significant improvement of our baseline in metrics such as minFDE6\mathrm{minFDE}_6 (as much as 20.3%) and MissRate6\mathrm{MissRate}_6 (as much as 33.3%), as well as a richer comprehension of map features demonstrated by different training configurations. The results obtained were consistent in all three data sets used for experiments: Argoverse, Interaction and NuScenes. We also submit our new pretrained model's results to the Interaction challenge and achieve 1st\textit{1st} place with respect to minFDE6\mathrm{minFDE}_6 and minADE6\mathrm{minADE}_6.

Keywords

Cite

@article{arxiv.2210.04672,
  title  = {Exploiting map information for self-supervised learning in motion forecasting},
  author = {Caio Azevedo and Thomas Gilles and Stefano Sabatini and Dzmitry Tsishkou},
  journal= {arXiv preprint arXiv:2210.04672},
  year   = {2022}
}
R2 v1 2026-06-28T03:09:00.065Z