The KITTI Vision Benchmark Suite

Method

FCN - Training with Large Patches [FTP]

Submitted on 11 Jan. 2016 14:24 by
Ankit Laddha (Carnegie Mellon University)

Running time:		0.28 s
Environment:		GPU @ 2.5 Ghz (C/C++)

Method Description:

Finetuning Deeplab FCN pretrained on Imagenet

Parameters:

-

Latex Bibtex:

@INPROCEEDINGS{Laddha2016IV,
author={Ankit Laddha and Mehmet Kemal Kocamaz and Luis E. Navarro-Serment and Martial Hebert},
booktitle={IEEE Intelligent Vehicles Symposium Proceedings},
title={Map-Supervised Road Detection},
year={2016},
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	91.20 %	90.60 %	91.11 %	91.29 %	4.06 %	8.71 %
UMM_ROAD	92.98 %	92.89 %	91.84 %	94.15 %	9.20 %	5.85 %
UU_ROAD	89.62 %	88.93 %	89.10 %	90.14 %	3.59 %	9.86 %
URBAN_ROAD	91.61 %	90.96 %	91.04 %	92.20 %	5.00 %	7.80 %

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