Method

Multimodal_RetinaNet [st] [fl] [la] [MMRetina]
[Anonymous Submission]

Submitted on 19 Nov. 2019 13:22 by
[Anonymous Submission]

Running time:0.38 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
Multimodal RetinaNet with RGB left camera image,
depth from stereo, optical flow from left camera
images and LIDAR Point Cloud as input data.
Running time computed without Depth, Optical Flow
and LIDAR Point Cloud Front View images extraction.
Parameters:
Backbone : Resnet18
Learning rate = 1e-4
Optimizer = Adam
Epochs = 400
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) 89.66 % 79.53 % 69.52 %
Pedestrian (Detection) 59.63 % 41.63 % 36.97 %
Cyclist (Detection) 43.71 % 28.00 % 24.62 %
This table as LaTeX


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



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



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




eXTReMe Tracker