EUVIMEDEUVIMED entdecken
Crossref

RMT-PPAD: Real-Time Multi-Task Learning for Panoptic Perception in Autonomous Driving

Jiayuan Wang, Q. M. Jonathan Wu, Ning Zhang, Katsuya Suto

IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 23253-23267

DOI: 10.1109/tmc.2026.3725133 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Jiayuan Wang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 43 Literaturangaben. Gefördert von: Canada Research Chairs Program, NSERC’s CREATE program on TrustCAV, JST ASPIRE.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Autonomous driving systems rely on panoptic driving perception that requires both precision and real-time performance. In this work, we propose RMT-PPAD, a real-time Transformer-based multi-task model that jointly performs object detection, drivable area segmentation, and lane line segmentation. Specifically, we introduce a lightweight gate control with an adapter module, which extracts branch-specific features for the detection and segmentation branches, while the gate adaptively fuses them with shared features to alleviate negative transfer. Additionally, we design a unified adaptive segmentation decoder to automatically learn task-specific weights for multi-scale features during training, which avoids the manual design of task-specific segmentation structures for different segmentation tasks. Furthermore, we identify the mismatch between the lane line label widths used for training and testing and address it by introducing dilated labels for fair evaluation. Experiments on the BDD100 K dataset show that RMT-PPAD achieves leading performance among the compared open-source MTL baselines, with mAP50 of 84.9% and Recall of 95.4% for object detection, mIoU of 92.6% for drivable area segmentation, and IoU of 56.8%, $F_{1}$ F 1 score of 66.5%, BFScore of 81.4%, and ACC of 84.7% for lane line segmentation under dilated labels. The inference speed reaches 32.6 FPS. Moreover, we introduce real-world scenarios to evaluate RMT-PPAD, where qualitative results show consistent perception outputs across diverse driving conditions.

Förderung

  • Canada Research Chairs Program
  • NSERC’s CREATE program on TrustCAV
  • JST ASPIRE

Zitieren

APA

Wang, J., Wu, Q. M. J., Zhang, N., & Suto, K. (2026). RMT-PPAD: Real-Time Multi-Task Learning for Panoptic Perception in Autonomous Driving. IEEE Transactions on Mobile Computing, 25(12), 23253-23267. https://doi.org/10.1109/tmc.2026.3725133

Vancouver

Wang J, Wu QMJ, Zhang N, Suto K. RMT-PPAD: Real-Time Multi-Task Learning for Panoptic Perception in Autonomous Driving. IEEE Transactions on Mobile Computing. 2026;25(12):23253-23267. doi:10.1109/tmc.2026.3725133

BibTeX
@article{Wang2026,
  author = {Wang, Jiayuan and Wu, Q. M. Jonathan and Zhang, Ning and Suto, Katsuya},
  title = {RMT-PPAD: Real-Time Multi-Task Learning for Panoptic Perception in Autonomous Driving},
  journal = {IEEE Transactions on Mobile Computing},
  year = {2026},
  volume = {25},
  number = {12},
  pages = {23253--23267},
  doi = {10.1109/tmc.2026.3725133}
}
Kostenlos starten →