The KITTI Vision Benchmark Suite

Method

RoadNet Real-Time [RoadNet-RT]

Submitted on 16 Aug. 2020 15:23 by
Lin Bai (Worcester Polytechnic Institute)

Running time:		8m s
Environment:		GPU @ 2.5 Ghz (Python)

Method Description:

two branches with spatial and context information
respectively

Parameters:

Amda: \alpha=1e-3
reduce on plateau
early stopping

Latex Bibtex:

@article{bai2020roadnet,
title={RoadNet-RT: High Throughput CNN
Architecture and SoC Design for Real-Time Road
Segmentation},
author={Bai, Lin and Lyu, Yecheng and Huang,
Xinming},
journal={arXiv preprint arXiv:2006.07644},
year={2020}
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	91.99 %	92.54 %	92.75 %	91.24 %	3.25 %	8.76 %
UMM_ROAD	93.98 %	95.19 %	94.47 %	93.49 %	6.01 %	6.51 %
UU_ROAD	90.79 %	91.67 %	91.79 %	89.80 %	2.62 %	10.20 %
URBAN_ROAD	92.55 %	93.21 %	92.94 %	92.16 %	3.86 %	7.84 %

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