The KITTI Vision Benchmark Suite

Method

HA-DeepLabv3+ [HA-DeepLabv3+]

Submitted on 18 Feb. 2021 18:15 by
Rui Fan (Tongji University)

Running time:		0.06 s
Environment:		1 core @ 2.5 Ghz (Python)

Method Description:

DeepLabv3+ trained on an augmented training set.
The training set was augmented using our proposed
DS-Generator.

Parameters:

identical to DeepLabv3+

Latex Bibtex:

@article{fan2020tmech,
title={Learning collision-free space detection
from stereo images: Homography matrix brings
better data augmentation},
author={Fan, Rui and Wang, Hengli and Cai,
Peide and Wu, Jin and Bocus, Mohammud Junaid and
Qiao, Lei and Liu, Ming},
journal={IEEE/ASME Transactions on
Mechatronics},
year={2021}
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	94.38 %	92.72 %	94.70 %	94.06 %	2.40 %	5.94 %
UMM_ROAD	96.10 %	95.03 %	95.48 %	96.73 %	5.03 %	3.27 %
UU_ROAD	93.24 %	91.83 %	93.19 %	93.28 %	2.22 %	6.72 %
URBAN_ROAD	94.83 %	93.24 %	94.77 %	94.89 %	2.88 %	5.11 %

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