makallouch / asayar

ASAYAR: A dataset for French and Arabic Text Detection in Highway panels and Traffic Sign Detection.

ASAYAR

This is a description for the paper:
ASAYAR: A Dataset for Arabic-Latin Scene Text Localization in Highway Traffic Panels
Mohammed Akallouch; Kaoutar Sefrioui Boujemaa; Afaf Bouhoute; Khalid Fardousse; Ismail Berrada

Overview

ASAYAR is the first public dataset dedicated for Latin (French) and Arabic Scene Text Detection in Highway panels. It comprises more than 1800 well-annotated images. The dataset was colleted from Moroccan Highway and it has been manually annotated. ASAYAR data can be used to develop and evaluate traffic signs detection and French or Arabic text detection in different languages.

Annotation format

In the dataset, each instance's location is annotated by a rectangle bounding boxes. The bounding box can be denoted as :
{XMIN, YMIN, XMAX, YMAX}. An obejct has a class name denoted as CLASS. The global image information are defined as follows: FOLDER, PATH, NAME, and SIZE. The following script shows an example of annotation :

<annotation>
    <folder>FOLDER</folder>
    <filename>IMAGE_NAME</filename>
    <path>PATH</path>
    <source>
        <database>ASAYAR</database>
    </source>
    <size>
        <width>WIDTH</width>
        <height>HEIGHT</height>
        <depth>DEPTH</depth>
    </size>
    <object>
        <name>CLASS</name>
        <bndbox>
            <xmin>XMIN</xmin>
            <ymin>YMIN</ymin>
            <xmax>XMAX</xmax>
            <ymax>YMAX</ymax>
        </bndbox>
    </object>
    ...
</annotation>

Dataset structure

Train or Test/
├── Traffic signs/
│   ├── Annotations/
│   │   ├── image_1.xml
│   │   └── ...
│   └── Images
│       ├── image_1.png
│       └── ...
│       
├── Text/
│   ├── Word Level/
│   │   ├── Annotations/
│   │   │   ├── image_1.xml
│   │   │   └── ...
│   │   └── Images/
│   │       ├── image_1.png
│   │       └── ...
│   └── Line Level/
│       ├── Annotations/
│       │   ├── image_1.xml
│       │   └── ...
│       └── Images/
│           ├── image_1.png
│           └── ...
└── Directional Symbols/
    ├── Annotations/
    │   ├── image_1.xml
    │   └── ...
    └── Images/
        ├── image_1.png
        └── ...

Import data

We provide a Jupyter Notebook with an example to import images and their annotations.

Convert to text format

To convert annotations from Voc pascal to txt format (xmin,ymin,xmax,ymax,class) use convert2txt.py.

Examples of Annotated Images

Donwload

The images and their annotations are available here. Download Link

Citation

Our paper introducing the dataset and the evaluations methods is published at the IEEE Transactions on Intelligent Transportation Systems 2020 and available here. If you make use of the ASAYAR dataset, please cite our following paper:

@ARTICLE{9233923,
      author={M. {Akallouch} and K. S. {Boujemaa} and A. {Bouhoute} and K. {Fardousse} and I. {Berrada}},
      journal={IEEE Transactions on Intelligent Transportation Systems}, 
      title={ASAYAR: A Dataset for Arabic-Latin Scene Text Localization in Highway Traffic Panels}, 
      year={2020},
      pages={1-11},
      doi={10.1109/TITS.2020.3029451}}