tanganke/gtsrb

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

Uses

Direct Use

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:

@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

[email protected]