Method

DetekTier3D [DT3D]
[Anonymous Submission]

Submitted on 29 Jul. 2018 14:52 by
[Anonymous Submission]

Running time:0,21s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
Meta-Architecture: Faster-RCNN
Feature Extractor: ResNet50
Inputdata: Densified depth image on rgb size
Parameters:
NVIDIA GeForce 1080 Ti (tensorflow-gpu)
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) 49.23 % 35.98 % 31.78 %
Car (3D Detection) 15.37 % 9.92 % 9.26 %
Car (Bird's Eye View) 23.38 % 17.19 % 13.86 %
Pedestrian (Detection) 27.02 % 19.19 % 18.98 %
Pedestrian (3D Detection) 1.14 % 1.14 % 1.14 %
Pedestrian (Bird's Eye View) 1.15 % 1.22 % 1.14 %
Cyclist (Detection) 31.29 % 20.65 % 20.73 %
Cyclist (3D Detection) 1.76 % 1.20 % 1.26 %
Cyclist (Bird's Eye View) 2.54 % 1.26 % 1.47 %
This table as LaTeX


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



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



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



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



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



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



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



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



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




eXTReMe Tracker