The KITTI Vision Benchmark Suite

Method

SNE-RoadSeg [SNE-RoadSeg]
https://sites.google.com/view/sne-roadseg/

Submitted on 10 Feb. 2020 11:16 by
Rui Fan (Tongji University)

Running time:		0.18 s
Environment:		GPU @ 2.5 Ghz (Python)

Method Description:

See https://github.com/hlwang1124/SNE-RoadSeg

Parameters:

See https://github.com/hlwang1124/SNE-RoadSeg

Latex Bibtex:

@InProceedings{fan2020sne-roadseg,
author = {Fan, Rui* and Wang, Hengli*
and Cai, Peide and Liu, Ming},
title = {SNE-RoadSeg: Incorporating
Surface Normal Information into Semantic
Segmentation for Accurate Freespace Detection},
booktitle = {ECCV},
year = {2020},
organization = {Springer},
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	96.42 %	93.67 %	96.59 %	96.26 %	1.55 %	3.74 %
UMM_ROAD	97.47 %	95.63 %	97.32 %	97.61 %	2.96 %	2.39 %
UU_ROAD	96.03 %	93.03 %	96.22 %	95.83 %	1.23 %	4.17 %
URBAN_ROAD	96.75 %	94.07 %	96.90 %	96.61 %	1.70 %	3.39 %

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.

This figure as: png eps pdf

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