Iran Traffic Sign Dataset (ITSD)

The Iran Traffic Sign Dataset (ITSD) is a comprehensive and valuable collection of traffic sign images, gathered under real driving conditions across various Iranian cities throughout all seasons of 2023 and 2024 (1402-1403 in the Persian calendar). This dataset has been designed and compiled to support artificial intelligence research in the field of traffic sign detection and recognition.The dataset consists of 14,111 high-resolution images containing 19,718 labeled traffic sign instances across 118 distinct classes, with a minimum of 10 samples per class. Annotation was performed using the MakeSense tool, with the results provided in two standard formats: Pascal VOC (.xml) and YOLO (.txt). The images, saved in .jpg format, have frame rates of 30fps or 60fps and resolutions of 1920x1080 or 1920x936. The recording was conducted using various mobile phones (including Galaxy S23 Ultra, iPhone 12, Galaxy J9, and Redmi 12) at speeds of up to 110 kilometers per hour.The dataset structure is designed to include six pre-defined folds to facilitate cross-validation. To ease its use by researchers, a Python script in the form of a Jupyter Notebook named Created Dataset is provided. This script takes the dataset path, the class configuration file (data_labels.yaml), an output path, and the name of the fold designated for testing as inputs. It then automatically splits the dataset into train, validation, and test subsets and saves them in the specified location. The six data folds, the dataset creation script, and the label mapping file are all packaged within a compressed folder named ITSD-dataset. Furthermore, the trained model weights for YOLOv9, generated for each test fold during the cross-validation process, are published within the same repository and are ready for use in evaluation or further training.There is an additional folder named Main, which contains the trained YOLOv12 weights for all six folds, along with a Python script named main.py. This script serves as the baseline detection system. After execution, it reads images from the inputs folder and saves the detection results in the outputs folder. Also a Readme file containing the version of the libraries used.When using the Iran Traffic Sign Dataset (ITSD) in your research, please support its development and respect the creators' rights by providing appropriate citation and acknowledgment.<b>Dataset specifications</b><b>Name</b>Iran Traffic Sign Dataset<b>Number of images</b>14111<b>Number of classes</b>118<b>Total number of objects</b>19718<b>Minimum number of samples per</b><b> </b><b>class</b>10<b>Annotation tool</b>Make-Sence<b>Annotation format</b>Voc pascal(.xml) وYolo (.txt)<b>Image size</b>1080*1920 , 936*1920<b>Image format</b>.jpg<b>Filming rate</b>30fps,60fps<b>Place of collection</b>All city s Iran<b>Collection season</b>All seasons<b>Accumulated time during the day</b>Full time<b>Number of folds</b>6 fold<b>Maximum video recording speed</b>110 km/h<b>Filming tool</b>Mobile(s23ultera,iphon12,GalaxyJ9,Redme12)<br>

伊朗交通标志数据集（Iran Traffic Sign Dataset, ITSD）是一套全面且极具价值的交通标志图像集合，采集自伊朗多座城市的真实驾驶场景，涵盖2023年至2024年（波斯历1402至1403年）的所有季节。本数据集的构建与汇编旨在为交通标志检测与识别领域的人工智能研究提供支撑。
数据集共包含14111张高分辨率图像，涵盖118个不同类别，总计19718个标注交通标志实例，每个类别至少包含10个样本。标注工作采用MakeSense工具完成，标注结果以两种标准格式输出：Pascal VOC（.xml）与YOLO（.txt）。图像均保存为.jpg格式，帧率为30fps或60fps，分辨率包含1920×1080与1920×936两种规格。数据采集使用多款移动设备完成，包括Galaxy S23 Ultra、iPhone 12、Galaxy J9与Redmi 12，采集时的最高行驶速度可达110公里每小时。
数据集结构预设6个划分折，以方便开展交叉验证研究。为便于研究人员使用，本数据集附带一份名为Created Dataset的Jupyter Notebook格式Python脚本。该脚本支持以数据集路径、类别配置文件（data_labels.yaml）、输出路径以及指定测试折的名称作为输入参数，自动将数据集划分为训练集、验证集与测试集，并将划分结果保存至指定路径。6个数据折、数据集构建脚本与标签映射文件均打包至名为ITSD-dataset的压缩文件夹中。
此外，交叉验证过程中为每个测试折生成的YOLOv9预训练模型权重，已发布至同一仓库中，可直接用于模型评估或后续微调训练。另有一个名为Main的附加文件夹，内含针对全部6个数据折训练得到的YOLOv12模型权重，以及一份名为main.py的Python脚本。该脚本为基线检测系统，运行后将从inputs文件夹读取图像，并将检测结果保存至outputs文件夹。文件夹中还包含一份记录了所使用库版本的Readme文件。
若在研究中使用伊朗交通标志数据集（ITSD），请通过恰当的引用与致谢方式支持其开发工作，并尊重创作者的知识产权。
<b>数据集规格参数</b>
<b>数据集名称</b>伊朗交通标志数据集
<b>图像总数</b>14111
<b>类别总数</b>118
<b>标注实例总数</b>19718
<b>单类别最小样本数</b>10
<b>标注工具</b>Make-Sence
<b>标注格式</b>Pascal VOC（.xml）与YOLO（.txt）
<b>图像分辨率</b>1920×1080、1920×936
<b>图像格式</b>.jpg
<b>采集帧率</b>30fps、60fps
<b>采集地点</b>伊朗全境各城市
<b>采集季节</b>全季节
<b>单日采集时长</b>全天
<b>划分折数量</b>6折
<b>最高采集行驶速度</b>110 km/h
<b>采集设备</b>移动设备（Galaxy S23 Ultra、iPhone 12、Galaxy J9、Redmi 12）