HRFuser: A Multi-resolution Sensor Fusion Architecture for 2D Object Detection

30 Jun 2022  ยท  Tim Broedermann, Christos Sakaridis, Dengxin Dai, Luc van Gool ยท

Besides standard cameras, autonomous vehicles typically include multiple additional sensors, such as lidars and radars, which help acquire richer information for perceiving the content of the driving scene. While several recent works focus on fusing certain pairs of sensors - such as camera with lidar or radar - by using architectural components specific to the examined setting, a generic and modular sensor fusion architecture is missing from the literature. In this work, we propose HRFuser, a modular architecture for multi-modal 2D object detection. It fuses multiple sensors in a multi-resolution fashion and scales to an arbitrary number of input modalities. The design of HRFuser is based on state-of-the-art high-resolution networks for image-only dense prediction and incorporates a novel multi-window cross-attention block as the means to perform fusion of multiple modalities at multiple resolutions. We demonstrate via extensive experiments on nuScenes and the adverse conditions DENSE datasets that our model effectively leverages complementary features from additional modalities, substantially improving upon camera-only performance and consistently outperforming state-of-the-art 3D and 2D fusion methods evaluated on 2D object detection metrics. The source code is publicly available.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
2D Object Detection Clear Weather HRFuser-T clear hard (AP) 79.48 # 1
Semantic Segmentation DeLiVER HRFuser (RGB) mIoU 47.95 # 11
Semantic Segmentation DeLiVER HRFuser (RGB-Depth) mIoU 51.88 # 9
Semantic Segmentation DeLiVER HRFuser (RGB-Event) mIoU 42.22 # 14
Semantic Segmentation DeLiVER HRFuser (RGB-LiDAR) mIoU 43.13 # 13
Semantic Segmentation DeLiVER HRFuser (RGB-D-Event) mIoU 51.83 # 10
Semantic Segmentation DeLiVER HRFuser (RGB-D-E-Li) mIoU 52.97 # 7
Semantic Segmentation DeLiVER HRFuser (RGB-D-LiDAR) mIoU 52.72 # 8
2D Object Detection Dense Fog HRFuser-T light fog hard (AP) 86.5 # 1
dense fog hard (AP) 78.21 # 1
snow/rain hard (AP) 78.09 # 1
Semantic Segmentation KITTI-360 HRFuser (RGB-D-LiDAR) mIoU 52.61 # 11
Semantic Segmentation KITTI-360 HRFuser (RGB-LiDAR) mIoU 48.74 # 13
Semantic Segmentation KITTI-360 HRFuser (RGB-Depth) mIoU 49.32 # 12

Methods


No methods listed for this paper. Add relevant methods here