vaclavpechtor/rvl_cdip-small-200

RVL-CDIP Small-200 Dataset

Dataset Summary

This is a subset of the RVL-CDIP (Ryerson Vision Lab Complex Document Information Processing) dataset, containing 200 samples per class for a total of 3,200 samples. The dataset consists of scanned document images in TIFF format, collected from various sources. The documents belong to 16 different categories, such as letter, memo, email, and more. The purpose of this dataset is to facilitate document classification tasks using NLP and computer vision techniques.

Supported Tasks and Leaderboards

• Document Classification: This dataset can be used for document classification tasks where the goal is to predict the correct category for each document image. No specific leaderboard is associated with this dataset.

Languages

The dataset contains documents in English.

Dataset Structure

Data Instances

A data instance consists of a TIFF image file representing a scanned document and its corresponding label indicating the document category.

Data Fields

• image: A TIFF image file representing a scanned document.
• label: A string representing the category of the document (e.g., "letter", "memo", "email", etc.).

Data Splits

The dataset is split into two subsets:

• Training set: Contains 200 samples per class, totaling 3,200 samples.
• Validation set: Contains a smaller number of samples per class.

Dataset Creation

Curation Rationale

This subset of the RVL-CDIP dataset was created to provide a smaller and more manageable dataset for researchers and practitioners who want to experiment with document classification tasks without the computational overhead of the full dataset.

Source Data

The dataset is a subset of the RVL-CDIP dataset, which contains 400,000 grayscale images in 16 classes, with 25,000 images per class.

Annotations

The dataset labels were derived from the original RVL-CDIP dataset. Each image file is associated with a label indicating its document category.

Personal and Sensitive Information

The dataset may contain personal or sensitive information, such as names, addresses, phone numbers, or email addresses. Users should take this into consideration when using the dataset.

Considerations for Using the Data

Social Impact of Dataset

This dataset can be used to develop models for document classification tasks, which can benefit a wide range of applications, such as document management systems, content analysis, and information retrieval.

Discussion of Biases

The dataset may contain biases due to the limited number of samples per class and the fact that the documents are sourced from different domains. These biases may affect the generalizability of models trained on this dataset.

Other Known Limitations

As this dataset is a small subset of the RVL-CDIP dataset, it may not be as representative or diverse as the full dataset. Additionally, the dataset only contains English documents, which may limit its applicability to other languages.

Additional Information

Licensing

Please refer to the RVL-CDIP dataset website for information on licensing and usage restrictions.

Citation Information

If you use this dataset, please cite the following paper: @inproceedings{harley2015evaluation, title={An evaluation of deep learning techniques for document image classification}, author={Harley, Adam W and Ufkes, Alex and Derpanis, Konstantinos G}, booktitle={2015 13th International Conference on Document Analysis and Recognition (ICDAR)}, pages={991--995}, year={2015}, organization={IEEE} }

Contact Information

For questions regarding the dataset, please refer to the RVL-CDIP dataset website for contact information.

Acknowledgements

This dataset is a subset of the RVL-CDIP dataset created by Adam W. Harley, Alex Ufkes, and Konstantinos G. Derpanis at the Ryerson Vision Lab (RVL), Ryerson University. The dataset creation was supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC).