Robust Collaborative 3D Object Detection in Presence of Pose Errors

Lu, Yifan; Li, Quanhao; Liu, Baoan; Dianati, Mehrdad; Chen, Feng; Chen, Siheng; Wang, Yanfeng

doi:10.1109/icra48891.2023.10160546

Cited by 26 publications

(7 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, we demonstrate the leading performance of DI-V2X through extensive experiments on the challenging DAIR-V2X (Yu et al 2022) and V2XSet (Xu et al 2022a) benchmark datasets. It significantly outperforms previous state-of-the-art (SOTA) models like CoAlign (Lu et al 2023) and V2X-ViT (Xu et al 2022a) by 5.76% AP@0.7 and 11.5% AP@0.7, respectively.…”

Section: Introductionmentioning

confidence: 82%

“…DAIR-V2X provides 3D object detection labels within the camera's field of view. We utilize these comprehensive 3D labels that encompass a 360-degree detection range, as mentioned by prior works (Lu et al 2023).…”

Section: Methodsmentioning

confidence: 99%

“…The proposed DMA randomly samples the instances from the database with the probability of 0.8, 0.1, and 0.1 for D f , D v , D i . We adopt the PointPillars (Lang et al 2019) as the detector to align with other V2X approaches (Lu et al 2023), while our approach is versatile to any BEV-based 3D object detectors. The feature map after fusion (i.e.,B i , B v ) is of size 252 × 100 × 256.…”

Section: Methodsmentioning

confidence: 99%

“…Where2comm (Hu et al 2022) focuses on the compression of feature maps, which only keep spatially sparse yet perceptually critical information. CoAlgin (Lu et al 2023) proposes an agent-object graph optimization method to handle the pose error. V2X-ViT (Xu et al 2022a) uses a transformer-based structure and a heterogeneous multiagent self-attention mechanism to facilitate feature fusion.…”

Section: V2x Collaborative Perceptionmentioning

confidence: 99%

“…Vehicle-infrastructure collaborative perception (Hu et al 2022;Xu et al 2023b;Lu et al 2023), also known as Vehicle-to-Everything (V2X) communication, has recently emerged as a promising approach to achieve full automation. It adequately leverages the sensor data collected from diverse agents, i.e., the vehicles and road-side infrastructures, to precisely perceive the complex driving scene.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

DI-V2X: Learning Domain-Invariant Representation for Vehicle-Infrastructure Collaborative 3D Object Detection

Li,

Yin,

et al. 2024

AAAI

View full text Add to dashboard Cite

Vehicle-to-Everything (V2X) collaborative perception has recently gained significant attention due to its capability to enhance scene understanding by integrating information from various agents, e.g., vehicles, and infrastructure. However, current works often treat the information from each agent equally, ignoring the inherent domain gap caused by the utilization of different LiDAR sensors of each agent, thus leading to suboptimal performance. In this paper, we propose DI-V2X, that aims to learn Domain-Invariant representations through a new distillation framework to mitigate the domain discrepancy in the context of V2X 3D object detection. DI-V2X comprises three essential components: a domain-mixing instance augmentation (DMA) module, a progressive domain-invariant distillation (PDD) module, and a domain-adaptive fusion (DAF) module. Specifically, DMA builds a domain-mixing 3D instance bank for the teacher and student models during training, resulting in aligned data representation. Next, PDD encourages the student models from different domains to gradually learn a domain-invariant feature representation towards the teacher, where the overlapping regions between agents are employed as guidance to facilitate the distillation process. Furthermore, DAF closes the domain gap between the students by incorporating calibration-aware domain-adaptive attention. Extensive experiments on the challenging DAIR-V2X and V2XSet benchmark datasets demonstrate DI-V2X achieves remarkable performance, outperforming all the previous V2X models. Code is available at https://github.com/Serenos/DI-V2X.

show abstract