Created by Feng Xue

Official Matlab Implementation for "A Novel Multi-layer Framework for Tiny Obstacle Discovery", ICRA 2019

This repository contains the official Matlab implementation for "A Novel Multi-layer Framework for Tiny Obstacle Discovery". This paper has been accepted by IEEE International Conference on Robotics and Automation (ICRA) 2019. And the official Python/ROS implementation is coming soon.

Note: This version has been slightly improved on the basis of the original work, mainly including the following points:

• The training code is slightly changed, ROC performance is improved.
• For higher efficiency, the basic edge detection algorithm uses structured edge detection [1].
• The modules in the system are largely accelerated, although there is still a lot of room for improvement.
• To be consistent with other works in the community, in the instance-level evaluation, the IoU is defined as the Intersection over Union between predicted proposals and the ground truth bounding boxes, which can be found at ./evaluation/Func_evaluation_DR.m

If you find the paper or the code useful, please cite our paper:

@INPROCEEDINGS{Xue_ICRA_2019,
Author = {Feng Xue, Anlong Ming, Menghan Zhou and Yu Zhou},
booktitle={2019 International Conference on Robotics and Automation (ICRA)},
title={A Novel Multi-layer Framework for Tiny Obstacle Discovery},
year={2019},
}

You may download the original images from here and annotations from here. Then, you will have the following directory structure:

lostandfound
|_ leftImg8bit
|  |_ test
|  |  |_ 02_Hanns_Klemm_Str_44
|  |  |_ ...
|  |  |_ 15_Rechbergstr_Deckenpfronn
|  |_ train
|  |  |_ 01_Hanns_Klemm_Str_45
|  |  |_ ...
|  |  |_ 14_Otto_Lilienthal_Str_24
|_ gtCoarse 
|  |_ test
|  |  |_ 02_Hanns_Klemm_Str_44
|  |  |_ ...
|  |  |_ 15_Rechbergstr_Deckenpfronn
|  |_ train
|  |  |_ 01_Hanns_Klemm_Str_45
|  |  |_ ...
|  |  |_ 14_Otto_Lilienthal_Str_24

For testing the algorithm on lost and found dataset, you can run the script in MATLAB:

./test.m

For Training the algorithm on lost and found dataset, you can run the script in MATLAB:

./train.m

The evaluations are conducted at the end of test.m. And there are three scripts for evaluation:

./evaluation/Func_evaluation_DR.m
./evaluation/Func_evaluation_ROC.m
./evaluation/Func_evaluation_time.m

For visualization, to run the script:

./result/figuredraw.m

The performances are presented as follows:

[1] Dollar P , Zitnick C L . Fast Edge Detection Using Structured Forests[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 37(8):1558-1570.

[2] Zitnick C L , Dollar P . Edge Boxes: Locating Object Proposals from Edges[C]// European Conference on Computer Vision. Springer, Cham, 2014.

[3] Ma J , Ming A , Huang Z , et al. Object-Level Proposals[C]// IEEE International Conference on Computer Vision. IEEE, 2017.

[4] Pinggera P , Ramos S , Gehrig S , et al. Lost and Found: Detecting Small Road Hazards for Self-Driving Vehicles[C]// IEEE/RSJ International Conference on Intelligent Robots and Systems, 2016.

2019/12/24, we are applying for the license. The code will be released after we obtain the license.

2020/07/04, the Matlab implementation is released.