The KITTI Vision Benchmark Suite

Method

StixelNet [StixelNet]

Submitted on 5 May. 2015 12:40 by
Ethan Fetaya (Weizmann inst.)

Running time:		1 s
Environment:		GPU @ 2.5 Ghz (C/C++)

Method Description:

In our approach we reduce the task to a column-wise
regression problem. The regression is then solved
using a deep convolutional neural network (CNN). In
addition, we introduce a new loss function based on
a semi-discrete representation to train the
network.

Parameters:

Latex Bibtex:

@INPROCEEDINGS{Levi2015BMVC,
Author={Levi, Dan and Garnett, Noa and
Fetaya,Ethan},
booktitle={ 26TH British Machine Vision Conference
(BMVC)},
title={StixelNet: A Deep Convolutional Network for
Obstacle Detection and Road Segmentation.},
year={2015},
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	85.33 %	72.14 %	81.21 %	89.89 %	9.48 %	10.11 %
UMM_ROAD	93.26 %	87.15 %	90.63 %	96.06 %	10.92 %	3.94 %
UU_ROAD	86.06 %	72.05 %	82.61 %	89.82 %	6.16 %	10.18 %
URBAN_ROAD	89.12 %	81.23 %	85.80 %	92.71 %	8.45 %	7.29 %

This table as LaTeX

Behavior Evaluation

Benchmark	PRE-20	F1-20	HR-20	PRE-30	F1-30	HR-30	PRE-40	F1-40	HR-40

This table as LaTeX

Road/Lane Detection

The following plots show precision/recall curves for the bird's eye view evaluation.

Distance-dependent Behavior Evaluation

The following plots show the F1 score/Precision/Hitrate with respect to the longitudinal distance which has been used for evaluation.

Visualization of Results

The following images illustrate the performance of the method qualitatively on a couple of test images. We first show results in the perspective image, followed by evaluation in bird's eye view. Here, red denotes false negatives, blue areas correspond to false positives and green represents true positives.

This figure as: png