Method

FIRM-Net [FIRM-Net]


Submitted on 3 Apr. 2024 08:39 by
Dehai Yu (Beihang University)

Running time:0.07 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
2 plug-and-play modules are proposed to enhance
the performance of existing 3d object
detector:global feature information recovering
module(G-FIRM)and local feature information
refinement module(L-FIRM).G-FIRM aims to recover
the lost information during the down-sampling
procedure, while L-FIRM focus on refining
representations of local features of each
ROI(regions of interest).G-FIRM significantly
boosts mAP on small objects such as pedestrians
and cyclists,but slightly decreases accuracy on
cars. This defect is mitigated by L-FIRM that
follows.
Parameters:
lr=0.005
weight decay=0.01
batchsize=2
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) 96.29 % 95.35 % 92.68 %
Car (Orientation) 96.27 % 95.20 % 92.41 %
Car (3D Detection) 88.25 % 81.65 % 76.98 %
Car (Bird's Eye View) 92.56 % 89.18 % 86.33 %
Pedestrian (Detection) 72.65 % 62.50 % 59.84 %
Pedestrian (Orientation) 66.69 % 56.33 % 53.53 %
Pedestrian (3D Detection) 51.65 % 43.43 % 40.97 %
Pedestrian (Bird's Eye View) 55.63 % 48.37 % 45.97 %
Cyclist (Detection) 89.49 % 77.51 % 70.60 %
Cyclist (Orientation) 89.35 % 77.23 % 70.31 %
Cyclist (3D Detection) 81.53 % 65.72 % 58.03 %
Cyclist (Bird's Eye View) 82.99 % 69.09 % 62.48 %
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



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



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