rasd3 / 3D-Dual-Fusion

[Arxiv 2022] This is the official implementation of 3D Dual-Fusion: Dual-Domain Dual-Query Camera-LiDAR Fusion for 3D Object Detection

3D Dual-Fusion

This project is the official implementation of 3D Dual-Fusion: Dual-Domain Dual-Query Camera-LiDAR Fusion for 3D Object Detection. We added the proposed modules to CenterPoint and TransFusion in NuScenes dataset and Voxel-RCNN in KITTI dataset. built on mmdetection3d, OpenPCDet, and Det3d. If our project was helpful to you, please cite:

@article{kim20223ddf,
  title={3D Dual-Fusion: Dual-Domain Dual-Query Camera-LiDAR Fusion for 3D Object Detection},
  author={Kim, Yecheol and Park, Konyul and Kim, Minwook and Kum, Dongsuk and Choi, Jun Won},
  journal={arXiv preprint arXiv:2211.13529},
  year={2022}
}

Introduction

Fusing data from cameras and LiDAR sensors is an essential technique to achieve robust 3D object detection. One key challenge in camera-LiDAR fusion involves mitigating the large domain gap between the two sensors in terms of coordinates and data distribution when fusing their features. In this paper, we propose a novel camera-LiDAR fusion architecture called, 3D Dual-Fusion, which is designed to mitigate the gap between the feature representations of camera and LiDAR data. The proposed method fuses the features of the camera-view and 3D voxel-view domain and models their interactions through deformable attention. We redesign the transformer fusion en- coder to aggregate the information from the two domains. Two major changes include 1) dual query-based deformable attention to fuse the dual-domain features interactively and 2) 3D local self-attention to encode the voxel-domain queries prior to dual- query decoding. The results of an experimental evaluation show that the proposed camera-LiDAR fusion architecture achieved competitive performance on the KITTI and nuScenes datasets, with state-of-the-art performances in some 3D object detection benchmark categories.

Main Results

KITTI Car 3D object detection validation

Model	Easy	Mod	Hard	Link
Voxel R-CNN	89.41	84.52	78.93	-
Voxel R-CNN + 3D-DF	92.80	85.96	83.62	TBD

NuScenes 3D object detection validation

3D Model	2D Model	mAP	NDS	Link
CenterPoint	-	59.6	66.8	-
CenterPoint + 3D-DF	DeepLabV3	67.3	71.1	Google
TransFusion-L	-	65.1	70.1	TBD
TransFusion-L + 3D-DF	ResNet50	69.3	72.2	Google
TransFusion-L + 3D-DF	Swin-Tiny	70.6	72.9	TBD

Acknowledgement

We sincerely thank the authors of FocalsConv, TransFusion, mmdetection3d, det3d, and OpenPCDet.