Method

YOLOv2 [YOLOv2]
https://github.com/cory8249/yolo2-pytorch

Submitted on 13 May. 2017 03:04 by
Cory Lee (Viterbi)

Running time:0.03 s
Environment:GPU @ 2.0 Ghz (Python + C/C++)

Method Description:
3rd party implementation of YOLOv2
Use pre-trained ConvNet on ImageNet
Train on KITTI for 100 epoch

https://arxiv.org/abs/1612.08242

* This is a simple experiment, not yet optimized
for KITTI.
Parameters:
Input Size = 1216 x 352
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) 86.40 % 69.01 % 59.57 %
Car (Orientation) 34.61 % 26.98 % 23.42 %
Pedestrian (Detection) 53.02 % 43.33 % 35.41 %
Pedestrian (Orientation) 32.98 % 27.35 % 22.99 %
Cyclist (Detection) 56.59 % 39.96 % 33.06 %
Cyclist (Orientation) 28.97 % 22.36 % 19.45 %
This table as LaTeX


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



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



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



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



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



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




eXTReMe Tracker