Unsupervised Domain Adaptation with Multimodal Fusion for Monocular 3D Object Detection
Vehicles · 01.05.2026 · 8 (5) · 98
DOI: 10.3390/vehicles8050098 ↗Kurz gefasst
Fachartikel in der Zeitschrift Vehicles, online erschienen am 01.05.2026. Verfasst von Jin Jiang und 6 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Major Science and Technology Special Project of Fujian Province, Applied Technology Engineering Center of Fujian Provincial Higher Education for Visual Perception and Intelligent Analysis.
Abstract (Originalsprache)
This paper presents UM3D, an end-to-end unsupervised domain adaptation framework for monocular 3D object detection. Monocular 3D object detection is appealing due to its low cost, yet it suffers from limited depth cues and poor cross-domain generalization when labeled data are scarce. Existing Pseudo-LiDAR methods require supervised training and propagate depth estimation errors to downstream detection, while current unsupervised domain adaptation (UDA) approaches exploit only a single modality and lack effective pseudo-label quality control. UM3D addresses these limitations through two key designs: (1) a quality-aware pseudo-label generation strategy with object-level random scaling and a memory bank refinement mechanism; and (2) an end-to-end differentiable pipeline that integrates multimodal fusion of image and Pseudo-LiDAR features with a multi-network consistency loss, which jointly optimizes depth estimation and 3D detection via backpropagation. Notably, the entire pipeline requires only a single monocular camera at inference; the Pseudo-LiDAR representation is generated internally from the same image, and thus the multimodal fusion integrates image and Pseudo-LiDAR features without requiring additional sensors. Extensive experiments across KITTI, nuScenes, Waymo, and Lyft demonstrate that UM3D generally outperforms existing UDA methods. In particular, a 19.30% relative APBEV improvement is achieved under easy conditions through end-to-end joint training compared to independent depth estimation, and up to 76.81% of the domain gap is closed on the WOD → KITTI benchmark.
Förderung
- Major Science and Technology Special Project of Fujian Province
- Applied Technology Engineering Center of Fujian Provincial Higher Education for Visual Perception and Intelligent Analysis
Zitieren
APA
Jiang, J., Dai, J., Li, W., Zhou, Y., Ye, M., Zhang, J., & Zhang, C. (2026). Unsupervised Domain Adaptation with Multimodal Fusion for Monocular 3D Object Detection. Vehicles, 8(5), 98. https://doi.org/10.3390/vehicles8050098
Vancouver
Jiang J, Dai J, Li W, Zhou Y, Ye M, Zhang J, et al. Unsupervised Domain Adaptation with Multimodal Fusion for Monocular 3D Object Detection. Vehicles. 2026;8(5):98. doi:10.3390/vehicles8050098
BibTeX
@article{Jiang2026,
author = {Jiang, Jin and Dai, Jidong and Li, Wei and Zhou, Yuquan and Ye, Maozhang and Zhang, Jianhuan and Zhang, Chentao},
title = {Unsupervised Domain Adaptation with Multimodal Fusion for Monocular 3D Object Detection},
journal = {Vehicles},
year = {2026},
volume = {8},
number = {5},
pages = {98},
doi = {10.3390/vehicles8050098}
}
EUVIMED