Abstract Accurate depth information is crucial for enhancing the performance of multi-view 3D object detection. Despite the success of some existing multi-view 3D detectors utilizing pixel-wise depth supervision, they overlook two significant phenomena: 1) the depth supervision obtained from LiDAR points is usually distributed on the surface of the object, which is not so friendly to existing DETR-based 3D detectors due to the lack of the depth of 3D object center; 2) for distant objects, fine-grained depth estimation of the whole object is more challenging. Therefore, we argue that the object-wise depth (or 3D center of the object) is essential for accurate detection. In this paper, we propose a new multi-view 3D object detector named OPEN, whose main idea is to effectively inject object-wise depth information into the network through our proposed object-wise position embedding. Specifically, we first employ an object-wise depth encoder, which takes the pixel-wise depth map as a prior, to accurately estimate the object-wise depth. Then, we utilize the proposed object-wise position embedding to encode the object-wise depth information into the transformer decoder, thereby producing 3D object-aware features for final detection. Extensive experiments verify the effectiveness of our proposed method. Furthermore, OPEN achieves a new state-of-the-art performance with 64.4% NDS and 56.7% mAP on the nuScenes test benchmark.

News

• 2024.07.02: Our another work SEED has also been accepted by ECCV 2024. 🎉
• 2024.07.02: OPEN has been accepted by ECCV 2024. 🎉

Results

• nuScenes Val Set

  The reproduced results are slightly higher than the reported results in the paper.

  R50：56.4 -> 56.5 NDS, 46.5 -> 47.0mAP

  R101: 60.6 -> 60.6 NDS, 51.6 -> 51.9 mAP

Model	Backbone	Pretrain	Resolution	NDS	mAP	Config	Download
OPEN	V2-99	DD3D	320 x 800	61.3	52.1	config	model
OPEN	R50	nuImage	256 x 704	56.5	47.0	config	model
OPEN	R101	nuImage	512 x 1408	60.6	51.9	config	model

• nuScenes Test Set

Model	Backbone	Pretrain	Resolution	NDS	mAP	Config	Download
OPEN	V2-99	DD3D	640 x 1600	64.4	56.7	config	model

TODO

• Release the paper.
• Release the code of OPEN.

Citation

@inproceedings{
  hou2024open,
  title={OPEN: Object-wise Position Embedding for Multi-view 3D Object Detection},
  author={Hou, Jinghua and Wang, Tong and Ye, Xiaoqing and Liu, Zhe and Tan, Xiao and Ding, Errui and Wang, Jingdong and Bai, Xiang},
  booktitle={ECCV},
  year={2024},
}

Acknowledgements

We thank these great works and open-source repositories: 3DPPE, StreamPETR, and MMDetection3D.