The KITTI Vision Benchmark Suite

Method

ACF-Net: Asymmetric Cascade Fusion for 3D Detection With LiDAR Point Clouds and Images [ACFNet]

Submitted on 17 Dec. 2023 07:54 by
YL Tian (University of Science and Technology of China)

Running time:		0.11 s
Environment:		1 core @ 2.5 Ghz (C/C++)

Method Description:

ACFNet adopts an asymmetric cascade fusion network to exploit both the structural information from point clouds and the complementary semantic information from images.

Parameters:

N/A

Latex Bibtex:

@ARTICLE{10363115,
author={Tian, Yonglin and Zhang, Xianjing and Wang, Xiao and Xu, Jintao and Wang, Jiangong and Ai, Rui and Gu, Weihao and Ding, Weiping},
journal={IEEE Transactions on Intelligent Vehicles},
title={ACF-Net: Asymmetric Cascade Fusion for 3D Detection With LiDAR Point Clouds and Images},
year={2023},
volume={},
number={},
pages={1-12},
doi={10.1109/TIV.2023.3341223}}

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).

Benchmark	Easy	Moderate	Hard
Car (Detection)	96.68 %	96.06 %	93.36 %
Car (Orientation)	96.64 %	95.95 %	93.17 %
Car (3D Detection)	90.80 %	84.67 %	80.14 %
Car (Bird's Eye View)	92.91 %	91.78 %	87.06 %
Pedestrian (Detection)	71.42 %	60.12 %	55.96 %
Pedestrian (Orientation)	65.55 %	53.97 %	49.97 %
Pedestrian (3D Detection)	54.62 %	46.36 %	42.57 %
Pedestrian (Bird's Eye View)	58.07 %	49.74 %	47.27 %
Cyclist (Detection)	86.92 %	76.15 %	71.33 %
Cyclist (Orientation)	86.11 %	75.34 %	70.41 %
Cyclist (3D Detection)	84.29 %	68.37 %	62.08 %
Cyclist (Bird's Eye View)	85.76 %	71.68 %	65.33 %