Skip to content

Latest commit

 

History

History
124 lines (106 loc) · 5.25 KB

README.md

File metadata and controls

124 lines (106 loc) · 5.25 KB

Annotations for Omnidirectional Pedestrian Detection

This page provides annotations for 4 publicly available datasets, Mirror Worlds Challenge, PIROPO, Bomni, and CVRG. The annotations are used in our paper, Omnidirectional Pedestrian Detection by Rotation Invariant Training. If you use our annotations, please follow a limitation, and cite our paper.

Description

In this page, only annotation files are provided. For training and evaluation, please download the original images from the original websites.

The annotation files for each sequence are placed in the same directory structure as the original, and the annotation file name is the same name as the original image file except for the extension. The annotations are provided in XML files that are based on Pascal VOC format. All the objects in the annotation files are named as "person".

We provide two types of annotations. One expresses pedestrians' area by horizontal bounding boxes, and the other expresses them by rotated bounding boxes. The former type of annotations are contained in "normal.tar.gz", and the latter type of annotations are contained in "rotate.tar.gz". When you use rotated version, you must rotate bounding boxes by yourself. Rotation angle for each bounding box is the angle between a vertical line and a line connecting a image center and bounding box center.

Annotations

Mirror Worlds challenge

The dataset can be downloaded from here.

The annotations are placed in "mw_18mar" directory in each tar file. Annotation statistics are as follows:

Sequence name #Annotation files #Instances
MW-18Mar-1 39 39
MW-18Mar-2 24 36
MW-18Mar-3 74 75
MW-18Mar-4 31 52
MW-18Mar-5 30 54
MW-18Mar-6 42 76
MW-18Mar-7 45 83
MW-18Mar-8 45 58
MW-18Mar-9 45 140
MW-18Mar-10 46 184
MW-18Mar-11 46 183
MW-18Mar-12 46 184
MW-18Mar-13 45 129
MW-18Mar-14 45 141
MW-18Mar-15 46 184
MW-18Mar-16 46 181
MW-18Mar-17 46 184
MW-18Mar-18 43 121
MW-18Mar-19 34 57
MW-18Mar-20 46 79
MW-18Mar-21 36 55
MW-18Mar-22 46 105
MW-18Mar-23 46 91
MW-18Mar-24 46 83
MW-18Mar-25 46 168
MW-18Mar-26 46 115
MW-18Mar-27 41 175
MW-18Mar-28 46 166
MW-18Mar-29 18 33
MW-18Mar-30 46 155
MW-18Mar-31 19 35

PIROPO

The dataset can be downloaded from here.

The annotations are placed in "piropo" directory in each tar file. Annotation statistics are as follows:

Sequence name #Annotation files #Instances
Room_A/omni_1A/omni_1A_training 750 750
Room_A/omni_1A/omni_1A_test2 119 296
Room_A/omni_1A/omni_1A_test3 141 141
Room_A/omni_2A/omni_2A_test3 120 120
Room_A/omni_2A/omni_2A_test2 113 261
Room_A/omni_2A/omni_2A_training 701 727
Room_A/omni_3A/omni_3A_test2 116 198
Room_A/omni_3A/omni_3A_training 487 487
Room_A/omni_3A/omni_3A_test3 65 65
Room_B/omni_1B/omni_1B_test2 27 48
Room_B/omni_1B/omni_1B_training 419 426
Room_B/omni_1B/omni_1B_test3 31 31

Bomni

The dataset can be downloaded from here.

The original dataset provides videos, so we split them into images. The image file name is determined with four-digit serial number starting from 1. The annotations are placed in "bomni" directory in each tar file. Annotation statistics are as follows:

Sequence name #Annotation files #Instances
scenario1/top-1 80 232
scenario1/top-3 91 351
scenario1/top-2 65 251
scenario1/top-0 101 288

CVRG

The dataset can be downloaded from here. The dataset we used is Dataset 1 of 4) Omnidirectional and panoramic image dataset (with annotations) to be used for human and car detection.

The annotations are placed in "cvrg" directory in each tar file. Annotation statistics are as follows:

Sequence name #Annotation files #Instances
Human Set/omni 30 71

Limitation

All the original datasets are provided for research purpose only. We provide our annotations following the original limitations.

Citation

If you use our annotations, please cite our paper:

@inproceedings{masato_wacv2019,
author = {Tamura, Masato and Horiguchi, Shota and Murakami, Tomokazu},
title = {Omnidirectional Pedestrian Detection by Rotation Invariant Training},
booktitle={2019 IEEE Winter Conference on Applications of Computer Vision (WACV)},
year = {2019},
month = {January},
pages = {1989-1998},
}