Method

Hyperbolic Cosine Transformer for LiDAR 3D Object Detection [ChTR3D]


Submitted on 1 Aug. 2022 11:26 by
帆航 杨 (天津理工大学)

Running time:0.06 s
Environment:1 core @ 2.5 Ghz (Python + C/C++)

Method Description:
we propose a two-stage 3D object detection
framework hyperbolic cosine transformer (ChTR3D)
for LIDAR point clouds,it refines proposals by
using cosh-attention module to encode rich
contextual dependencies among points with a linear
computational complexity.ChTR3D generates proposal
at the first stage by using 3D Sparse Convolution
to extract features from voxels, and it refine
proposal at the second stage by applying the cosh-
attention-based encoder-decoder architecture to
extract features from point clouds with linear
calculation.
Parameters:
α=1.1
Latex Bibtex:

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) 96.32 % 94.95 % 90.25 %
Car (Orientation) 96.31 % 94.84 % 90.12 %
Car (3D Detection) 87.81 % 81.19 % 76.70 %
Car (Bird's Eye View) 92.58 % 88.85 % 85.98 %
This table as LaTeX

2D object detection results.
This figure as: png eps txt gnuplot



Orientation estimation results.
This figure as: png eps txt gnuplot



3D object detection results.
This figure as: png eps txt gnuplot



Bird's eye view results.
This figure as: png eps txt gnuplot




eXTReMe Tracker