five

tanganke/gtsrb

收藏
Hugging Face2024-05-07 更新2024-06-12 收录
下载链接:
https://hf-mirror.com/datasets/tanganke/gtsrb
下载链接
链接失效反馈
官方服务:
资源简介:
--- language: - en size_categories: - 10K<n<100K task_categories: - image-classification dataset_info: features: - name: image dtype: image - name: label dtype: class_label: names: '0': red and white circle 20 kph speed limit '1': red and white circle 30 kph speed limit '2': red and white circle 50 kph speed limit '3': red and white circle 60 kph speed limit '4': red and white circle 70 kph speed limit '5': red and white circle 80 kph speed limit '6': end / de-restriction of 80 kph speed limit '7': red and white circle 100 kph speed limit '8': red and white circle 120 kph speed limit '9': red and white circle red car and black car no passing '10': red and white circle red truck and black car no passing '11': red and white triangle road intersection warning '12': white and yellow diamond priority road '13': red and white upside down triangle yield right-of-way '14': stop '15': empty red and white circle '16': red and white circle no truck entry '17': red circle with white horizonal stripe no entry '18': red and white triangle with exclamation mark warning '19': red and white triangle with black left curve approaching warning '20': red and white triangle with black right curve approaching warning '21': red and white triangle with black double curve approaching warning '22': red and white triangle rough / bumpy road warning '23': red and white triangle car skidding / slipping warning '24': red and white triangle with merging / narrow lanes warning '25': red and white triangle with person digging / construction / road work warning '26': red and white triangle with traffic light approaching warning '27': red and white triangle with person walking warning '28': red and white triangle with child and person walking warning '29': red and white triangle with bicyle warning '30': red and white triangle with snowflake / ice warning '31': red and white triangle with deer warning '32': white circle with gray strike bar no speed limit '33': blue circle with white right turn arrow mandatory '34': blue circle with white left turn arrow mandatory '35': blue circle with white forward arrow mandatory '36': blue circle with white forward or right turn arrow mandatory '37': blue circle with white forward or left turn arrow mandatory '38': blue circle with white keep right arrow mandatory '39': blue circle with white keep left arrow mandatory '40': blue circle with white arrows indicating a traffic circle '41': white circle with gray strike bar indicating no passing for cars has ended '42': white circle with gray strike bar indicating no passing for trucks has ended splits: - name: train num_bytes: 252930879.36 num_examples: 26640 - name: test num_bytes: 104816357.02 num_examples: 12630 - name: contrast num_bytes: 104816357.02 num_examples: 12630 - name: gaussian_noise num_bytes: 104816357.02 num_examples: 12630 - name: impulse_noise num_bytes: 104816357.02 num_examples: 12630 - name: jpeg_compression num_bytes: 104816357.02 num_examples: 12630 - name: motion_blur num_bytes: 104816357.02 num_examples: 12630 - name: pixelate num_bytes: 39121740.4 num_examples: 12630 - name: spatter num_bytes: 104816357.02 num_examples: 12630 download_size: 1027074522 dataset_size: 1025767118.8999999 configs: - config_name: default data_files: - split: train path: data/train-* - split: test path: data/test-* - split: contrast path: data/contrast-* - split: gaussian_noise path: data/gaussian_noise-* - split: impulse_noise path: data/impulse_noise-* - split: jpeg_compression path: data/jpeg_compression-* - split: motion_blur path: data/motion_blur-* - split: pixelate path: data/pixelate-* - split: spatter path: data/spatter-* --- # Dataset Card for German Traffic Sign Recognition Benchmark This dataset contains images of 43 classes of traffic signs. It is intended for developing and benchmarking traffic sign recognition systems. ## Dataset Details ### Dataset Description The German Traffic Sign Recognition Benchmark (GTSRB) is a multi-class classification dataset featuring 43 classes of traffic signs. The images were cropped from a larger set of images to focus on the traffic sign and eliminate background. Multiple data augmentations such as Gaussian noise, motion blur, contrast changes, etc. are provided as additional test sets to benchmark model robustness. ### Dataset Sources - [Paper with code](https://paperswithcode.com/dataset/gtsrb) ## Uses ### Direct Use ```python from datasets import load_dataset dataset = load_dataset('tanganke/gtsrb') ``` ## Dataset Structure The dataset is provided in 9 splits, including training data and clean test data: - train: 26,640 images - test: 12,630 images and 7 kinds of corrupted test datasets to evaluate the robustness: - contrast: 12,630 contrast-adjusted test images - gaussian_noise: 12,630 Gaussian noise augmented test images - impulse_noise: 12,630 impulse noise augmented test images - jpeg_compression: 12,630 JPEG-compressed test images - motion_blur: 12,630 motion-blurred test images - pixelate: 12,630 pixelated test images - spatter: 12,630 spatter augmented test images Each split contains 43 classes of traffic signs, with the class labels and names specified in the dataset metadata. ## Citation [optional] You can use any of the provided BibTeX entries for your reference list: ```bibtex @article{stallkampManVsComputer2012, title = {Man vs. Computer: {{Benchmarking}} Machine Learning Algorithms for Traffic Sign Recognition}, shorttitle = {Man vs. Computer}, author = {Stallkamp, J. and Schlipsing, M. and Salmen, J. and Igel, C.}, year = {2012}, month = aug, journal = {Neural Networks}, series = {Selected {{Papers}} from {{IJCNN}} 2011}, volume = {32}, pages = {323--332}, issn = {0893-6080}, doi = {10.1016/j.neunet.2012.02.016}, url = {https://www.sciencedirect.com/science/article/pii/S0893608012000457}, keywords = {Benchmarking,Convolutional neural networks,Machine learning,Traffic sign recognition} } @misc{yangAdaMergingAdaptiveModel2023, title = {{{AdaMerging}}: {{Adaptive Model Merging}} for {{Multi-Task Learning}}}, shorttitle = {{{AdaMerging}}}, author = {Yang, Enneng and Wang, Zhenyi and Shen, Li and Liu, Shiwei and Guo, Guibing and Wang, Xingwei and Tao, Dacheng}, year = {2023}, month = oct, number = {arXiv:2310.02575}, eprint = {2310.02575}, primaryclass = {cs}, publisher = {arXiv}, doi = {10.48550/arXiv.2310.02575}, url = {http://arxiv.org/abs/2310.02575}, archiveprefix = {arxiv}, keywords = {Computer Science - Computer Vision and Pattern Recognition,Computer Science - Machine Learning} } @misc{tangConcreteSubspaceLearning2023, title = {Concrete {{Subspace Learning}} Based {{Interference Elimination}} for {{Multi-task Model Fusion}}}, author = {Tang, Anke and Shen, Li and Luo, Yong and Ding, Liang and Hu, Han and Du, Bo and Tao, Dacheng}, year = {2023}, month = dec, number = {arXiv:2312.06173}, eprint = {2312.06173}, publisher = {arXiv}, url = {http://arxiv.org/abs/2312.06173}, archiveprefix = {arxiv}, copyright = {All rights reserved}, keywords = {Computer Science - Machine Learning} } @misc{tangMergingMultiTaskModels2024, title = {Merging {{Multi-Task Models}} via {{Weight-Ensembling Mixture}} of {{Experts}}}, author = {Tang, Anke and Shen, Li and Luo, Yong and Yin, Nan and Zhang, Lefei and Tao, Dacheng}, year = {2024}, month = feb, number = {arXiv:2402.00433}, eprint = {2402.00433}, primaryclass = {cs}, publisher = {arXiv}, doi = {10.48550/arXiv.2402.00433}, url = {http://arxiv.org/abs/2402.00433}, archiveprefix = {arxiv}, copyright = {All rights reserved}, keywords = {Computer Science - Computer Vision and Pattern Recognition,Computer Science - Machine Learning} } ``` ## Dataset Card Authors Anke Tang ## Dataset Card Contact [tang.anke@foxmail.com](mailto:tang.anke@foxmail.com)

语言: - 英语 样本量范围: - 10000 < 样本数量 < 100000 任务类别: - 图像分类(image-classification) 数据集信息: 特征: - 名称:image 数据类型:图像 - 名称:label 数据类型: 类别标签: 类别名称: '0': 红白圆形20公里/小时限速标志 '1': 红白圆形30公里/小时限速标志 '2': 红白圆形50公里/小时限速标志 '3': 红白圆形60公里/小时限速标志 '4': 红白圆形70公里/小时限速标志 '5': 红白圆形80公里/小时限速标志 '6': 80公里/小时限速解除标志 '7': 红白圆形100公里/小时限速标志 '8': 红白圆形120公里/小时限速标志 '9': 红白圆形禁止超车(红车黑车图案)标志 '10': 红白圆形禁止货车超车(红货车黑车图案)标志 '11': 红白三角形道路交叉路口警告标志 '12': 白黄菱形优先通行道路标志 '13': 倒红白三角形优先通行权让行标志 '14': 停车让行标志 '15': 空白红白圆形标志 '16': 红白圆形禁止货车驶入标志 '17': 带白色横条的红色圆形禁止驶入标志 '18': 带感叹号的红白三角形通用警告标志 '19': 带黑色左侧弯道的红白三角形左侧弯道警告标志 '20': 带黑色右侧弯道的红白三角形右侧弯道警告标志 '21': 带黑色反向弯道的红白三角形反向弯道警告标志 '22': 红白三角形路面颠簸警告标志 '23': 红白三角形车辆侧滑警告标志 '24': 红白三角形车道收窄/合流警告标志 '25': 带施工人员图案的红白三角形道路施工警告标志 '26': 红白三角形前方信号灯警告标志 '27': 红白三角形前方行人通行警告标志 '28': 红白三角形前方儿童及行人通行警告标志 '29': 红白三角形前方非机动车(自行车)通行警告标志 '30': 红白三角形前方易结冰/积雪警告标志 '31': 红白三角形前方野生动物(鹿)横穿警告标志 '32': 带灰色横杠的白色圆形限速解除标志 '33': 蓝色圆形白色右转箭头强制通行标志 '34': 蓝色圆形白色左转箭头强制通行标志 '35': 蓝色圆形白色直行箭头强制通行标志 '36': 蓝色圆形白色直行或右转箭头强制通行标志 '37': 蓝色圆形白色直行或左转箭头强制通行标志 '38': 蓝色圆形白色靠右行驶箭头强制通行标志 '39': 蓝色圆形白色靠左行驶箭头强制通行标志 '40': 蓝色圆形白色环岛行驶箭头强制通行标志 '41': 带灰色横杠的白色圆形小型车辆禁止超车解除标志 '42': 带灰色横杠的白色圆形货车禁止超车解除标志 划分集: - 名称:train 字节数:252930879.36 样本数量:26640 - 名称:test 字节数:104816357.02 样本数量:12630 - 名称:contrast 字节数:104816357.02 样本数量:12630 - 名称:gaussian_noise 字节数:104816357.02 样本数量:12630 - 名称:impulse_noise 字节数:104816357.02 样本数量:12630 - 名称:jpeg_compression 字节数:104816357.02 样本数量:12630 - 名称:motion_blur 字节数:104816357.02 样本数量:12630 - 名称:pixelate 字节数:39121740.4 样本数量:12630 - 名称:spatter 字节数:104816357.02 样本数量:12630 下载大小:1027074522 数据集总大小:1025767118.8999999 配置: - 配置名称:default 数据文件: - 划分集:train 路径:data/train-* - 划分集:test 路径:data/test-* - 划分集:contrast 路径:data/contrast-* - 划分集:gaussian_noise 路径:data/gaussian_noise-* - 划分集:impulse_noise 路径:data/impulse_noise-* - 划分集:jpeg_compression 路径:data/jpeg_compression-* - 划分集:motion_blur 路径:data/motion_blur-* - 划分集:pixelate 路径:data/pixelate-* - 划分集:spatter 路径:data/spatter-* # 德国交通标志识别基准数据集卡片(German Traffic Sign Recognition Benchmark) 本数据集包含43类交通标志的图像,旨在用于开发和评测交通标志识别系统。 ## 数据集详情 ### 数据集描述 德国交通标志识别基准数据集(German Traffic Sign Recognition Benchmark, GTSRB)是一个包含43类交通标志的多分类数据集。 所有图像均从更大规模的图像集合中裁剪得到,仅保留交通标志主体以消除背景干扰。 为评测模型鲁棒性,本数据集额外提供了多种数据增强后的测试集,包括高斯噪声(Gaussian noise)、运动模糊(motion blur)、对比度调整等。 ### 数据集来源 - [论文与代码页面](https://paperswithcode.com/dataset/gtsrb) ## 直接使用示例 python from datasets import load_dataset dataset = load_dataset('tanganke/gtsrb') ## 数据集结构 本数据集共包含9个划分集,其中包括训练集与干净测试集: - train:26,640张图像 - test:12,630张图像 以及7种带干扰的测试数据集,用于评估模型鲁棒性: - contrast:12,630张对比度调整后的测试图像 - gaussian_noise:12,630张添加了高斯噪声(Gaussian noise)的测试图像 - impulse_noise:12,630张添加了脉冲噪声(impulse noise)的测试图像 - jpeg_compression:12,630张经过JPEG压缩(JPEG compression)的测试图像 - motion_blur:12,630张带有运动模糊(motion blur)的测试图像 - pixelate:12,630张经过像素化(pixelate)处理的测试图像 - spatter:12,630张添加了污渍(spatter)效果的测试图像 每个划分集均包含全部43类交通标志,类别标签与名称已在数据集元数据中定义。 ## 引用文献[可选] 可在参考文献中使用以下任意BibTeX条目: bibtex @article{stallkampManVsComputer2012, title = {Man vs. Computer: {{Benchmarking}} Machine Learning Algorithms for Traffic Sign Recognition}, shorttitle = {Man vs. Computer}, author = {Stallkamp, J. and Schlipsing, M. and Salmen, J. and Igel, C.}, year = {2012}, month = aug, journal = {Neural Networks}, series = {Selected {{Papers}} from {{IJCNN}} 2011}, volume = {32}, pages = {323--332}, issn = {0893-6080}, doi = {10.1016/j.neunet.2012.02.016}, url = {https://www.sciencedirect.com/science/article/pii/S0893608012000457}, keywords = {Benchmarking,Convolutional neural networks,Machine learning,Traffic sign recognition} } @misc{yangAdaMergingAdaptiveModel2023, title = {{{AdaMerging}}: {{Adaptive Model Merging}} for {{Multi-Task Learning}}}, shorttitle = {{{AdaMerging}}}, author = {Yang, Enneng and Wang, Zhenyi and Shen, Li and Liu, Shiwei and Guo, Guibing and Wang, Xingwei and Tao, Dacheng}, year = {2023}, month = oct, number = {arXiv:2310.02575}, eprint = {2310.02575}, primaryclass = {cs}, publisher = {arXiv}, doi = {10.48550/arXiv.2310.02575}, url = {http://arxiv.org/abs/2310.02575}, archiveprefix = {arxiv}, keywords = {Computer Science - Computer Vision and Pattern Recognition,Computer Science - Machine Learning} } @misc{tangConcreteSubspaceLearning2023, title = {Concrete {{Subspace Learning}} Based {{Interference Elimination}} for {{Multi-task Model Fusion}}}, author = {Tang, Anke and Shen, Li and Luo, Yong and Ding, Liang and Hu, Han and Du, Bo and Tao, Dacheng}, year = {2023}, month = dec, number = {arXiv:2312.06173}, eprint = {2312.06173}, publisher = {arXiv}, url = {http://arxiv.org/abs/2312.06173}, archiveprefix = {arxiv}, copyright = {All rights reserved}, keywords = {Computer Science - Machine Learning} } @misc{tangMergingMultiTaskModels2024, title = {Merging {{Multi-Task Models}} via {{Weight-Ensembling Mixture}} of {{Experts}}}, author = {Tang, Anke and Shen, Li and Luo, Yong and Yin, Nan and Zhang, Lefei and Tao, Dacheng}, year = {2024}, month = feb, number = {arXiv:2402.00433}, eprint = {2402.00433}, primaryclass = {cs}, publisher = {arXiv}, doi = {10.48550/arXiv.2402.00433}, url = {http://arxiv.org/abs/2402.00433}, archiveprefix = {arxiv}, copyright = {All rights reserved}, keywords = {Computer Science - Computer Vision and Pattern Recognition,Computer Science - Machine Learning} } ## 数据集卡片作者 Anke Tang ## 数据集卡片联系方式 [ tang.anke@foxmail.com ](mailto:tang.anke@foxmail.com)
提供机构:
tanganke
原始信息汇总

数据集概述

数据集基本信息

  • 语言: 英语
  • 大小: 10K<n<100K
  • 任务类型: 图像分类

数据集特征

  • 图像: 数据类型为图像
  • 标签: 数据类型为类别标签,包含43个类别,每个类别对应一个特定的交通标志描述。

数据集分割

  • 训练集: 26,640个样本
  • 测试集: 12,630个样本
  • 对比测试集: 12,630个样本
  • 高斯噪声测试集: 12,630个样本
  • 脉冲噪声测试集: 12,630个样本
  • JPEG压缩测试集: 12,630个样本
  • 运动模糊测试集: 12,630个样本
  • 像素化测试集: 12,630个样本
  • 喷溅测试集: 12,630个样本

数据集用途

  • 用于开发和基准测试交通标志识别系统。

数据集结构

  • 数据集包含9个分割,包括训练数据和多种数据增强后的测试数据,用于评估模型的鲁棒性。

数据集标签详情

  • 标签从0到42,每个标签对应一个具体的交通标志描述,如0: red and white circle 20 kph speed limit。
搜集汇总
数据集介绍
main_image_url
构建方式
tanganke/gtsrb数据集的构建,旨在为交通标志识别系统提供训练与评估的基准。该数据集通过裁剪原始图像以聚焦交通标志本身,并排除背景干扰。此外,通过引入多种数据增强技术,如高斯噪声、运动模糊、对比度调整等,以模拟实际场景中的图像变化,增强模型的泛化能力。
使用方法
使用tanganke/gtsrb数据集,研究者可以通过HuggingFace的datasets库直接加载。加载后,数据集被分为多个split,包括训练集、测试集以及多种增强的测试集,方便研究者进行模型的训练、验证和测试。每种split都包含完整的交通标志类别,可以直接用于机器学习模型的训练与评估。
背景与挑战
背景概述
德国交通标志识别基准(GTSRB)数据集是一款包含43类交通标志图像的多类分类数据集。该数据集的构建旨在为交通标志识别系统的研发与评估提供标准平台。该数据集起源于2012年,由Stallkamp等人创建,并发表于《Neural Networks》期刊。GTSRB数据集的核心研究问题是提高交通标志识别系统的准确性和鲁棒性,其对计算机视觉和机器学习领域产生了显著影响,成为评估相关算法性能的重要基准之一。
当前挑战
该数据集在构建过程中遇到的挑战主要包括如何有效提取图像特征、处理图像背景噪声以及增强模型对不同光照和天气条件的适应性。在研究领域问题上,GTSRB数据集面临的挑战包括识别准确性的提高、模型对于各类交通标志的泛化能力、以及对抗性样本的鲁棒性。数据集提供了多种图像增强版本,如对比度调整、高斯噪声、运动模糊等,用以评估模型在不同类型干扰下的表现,从而推动了相关领域的研究进展。
常用场景
经典使用场景
在智能交通系统的研究领域,tanganke/gtsrb数据集的经典使用场景在于交通标志的自动识别。该数据集包含了43种不同类型的交通标志图像,这些图像被用于训练和评估深度学习模型,以便能够准确识别道路上的各种交通标志,从而为自动驾驶车辆提供决策支持。
解决学术问题
该数据集解决了学术研究中如何提高交通标志识别准确率和模型鲁棒性的问题。通过提供多种数据增强版本,如对比度调整、高斯噪声、运动模糊等,研究者可以评估和改进模型在不同视觉条件下的性能,这对于自动驾驶技术的安全性和可靠性至关重要。
实际应用
在实际应用中,tanganke/gtsrb数据集的成果已被广泛应用于自动驾驶系统的开发中。通过准确识别交通标志,自动驾驶车辆能够更好地遵守交通规则,提高行驶安全性。此外,该数据集的成果还可用于交通监控和管理,辅助执法部门进行交通违法行为的自动识别。
数据集最近研究
最新研究方向
tanganke/gtsrb数据集作为交通标志识别的重要基准,近期研究集中于提高模型的鲁棒性和准确性。例如,研究者们探讨了AdaMerging自适应模型合并方法,以优化多任务学习;同时,基于具体子空间学习的方法被提出以消除多任务模型融合中的干扰。此外,权值组合的专家混合模型合并策略也受到关注,这些研究为交通标志识别系统在实际应用中的性能提升提供了新的视角和技术路径。
以上内容由遇见数据集搜集并总结生成
5,000+
优质数据集
54 个
任务类型
进入经典数据集
二维码
社区交流群

面向社区/商业的数据集话题

二维码
科研交流群

面向高校/科研机构的开源数据集话题

数据驱动未来

携手共赢发展

商业合作