aharley/rvl_cdip
收藏资源简介:
--- annotations_creators: - found language_creators: - found language: - en license: - other multilinguality: - monolingual size_categories: - 100K<n<1M source_datasets: - extended|iit_cdip task_categories: - image-classification task_ids: - multi-class-image-classification paperswithcode_id: rvl-cdip pretty_name: RVL-CDIP viewer: false dataset_info: features: - name: image dtype: image - name: label dtype: class_label: names: '0': letter '1': form '2': email '3': handwritten '4': advertisement '5': scientific report '6': scientific publication '7': specification '8': file folder '9': news article '10': budget '11': invoice '12': presentation '13': questionnaire '14': resume '15': memo splits: - name: train num_bytes: 38816373360 num_examples: 320000 - name: test num_bytes: 4863300853 num_examples: 40000 - name: validation num_bytes: 4868685208 num_examples: 40000 download_size: 38779484559 dataset_size: 48548359421 --- # Dataset Card for RVL-CDIP ## Table of Contents - [Dataset Description](#dataset-description) - [Dataset Summary](#dataset-summary) - [Supported Tasks](#supported-tasks-and-leaderboards) - [Languages](#languages) - [Dataset Structure](#dataset-structure) - [Data Instances](#data-instances) - [Data Fields](#data-instances) - [Data Splits](#data-instances) - [Dataset Creation](#dataset-creation) - [Curation Rationale](#curation-rationale) - [Source Data](#source-data) - [Annotations](#annotations) - [Personal and Sensitive Information](#personal-and-sensitive-information) - [Considerations for Using the Data](#considerations-for-using-the-data) - [Social Impact of Dataset](#social-impact-of-dataset) - [Discussion of Biases](#discussion-of-biases) - [Other Known Limitations](#other-known-limitations) - [Additional Information](#additional-information) - [Dataset Curators](#dataset-curators) - [Licensing Information](#licensing-information) - [Citation Information](#citation-information) ## Dataset Description - **Homepage:** [The RVL-CDIP Dataset](https://www.cs.cmu.edu/~aharley/rvl-cdip/) - **Repository:** - **Paper:** [Evaluation of Deep Convolutional Nets for Document Image Classification and Retrieval](https://arxiv.org/abs/1502.07058) - **Leaderboard:** [RVL-CDIP leaderboard](https://paperswithcode.com/dataset/rvl-cdip) - **Point of Contact:** [Adam W. Harley](mailto:aharley@cmu.edu) ### Dataset Summary The RVL-CDIP (Ryerson Vision Lab Complex Document Information Processing) dataset consists of 400,000 grayscale images in 16 classes, with 25,000 images per class. There are 320,000 training images, 40,000 validation images, and 40,000 test images. The images are sized so their largest dimension does not exceed 1000 pixels. ### Supported Tasks and Leaderboards - `image-classification`: The goal of this task is to classify a given document into one of 16 classes representing document types (letter, form, etc.). The leaderboard for this task is available [here](https://paperswithcode.com/sota/document-image-classification-on-rvl-cdip). ### Languages All the classes and documents use English as their primary language. ## Dataset Structure ### Data Instances A sample from the training set is provided below : ``` { 'image': <PIL.TiffImagePlugin.TiffImageFile image mode=L size=754x1000 at 0x7F9A5E92CA90>, 'label': 15 } ``` ### Data Fields - `image`: A `PIL.Image.Image` object containing a document. - `label`: an `int` classification label. <details> <summary>Class Label Mappings</summary> ```json { "0": "letter", "1": "form", "2": "email", "3": "handwritten", "4": "advertisement", "5": "scientific report", "6": "scientific publication", "7": "specification", "8": "file folder", "9": "news article", "10": "budget", "11": "invoice", "12": "presentation", "13": "questionnaire", "14": "resume", "15": "memo" } ``` </details> ### Data Splits | |train|test|validation| |----------|----:|----:|---------:| |# of examples|320000|40000|40000| The dataset was split in proportions similar to those of ImageNet. - 320000 images were used for training, - 40000 images for validation, and - 40000 images for testing. ## Dataset Creation ### Curation Rationale From the paper: > This work makes available a new labelled subset of the IIT-CDIP collection, containing 400,000 document images across 16 categories, useful for training new CNNs for document analysis. ### Source Data #### Initial Data Collection and Normalization The same as in the IIT-CDIP collection. #### Who are the source language producers? The same as in the IIT-CDIP collection. ### Annotations #### Annotation process The same as in the IIT-CDIP collection. #### Who are the annotators? The same as in the IIT-CDIP collection. ### Personal and Sensitive Information [More Information Needed] ## Considerations for Using the Data ### Social Impact of Dataset [More Information Needed] ### Discussion of Biases [More Information Needed] ### Other Known Limitations [More Information Needed] ## Additional Information ### Dataset Curators The dataset was curated by the authors - Adam W. Harley, Alex Ufkes, and Konstantinos G. Derpanis. ### Licensing Information RVL-CDIP is a subset of IIT-CDIP, which came from the [Legacy Tobacco Document Library](https://www.industrydocuments.ucsf.edu/tobacco/), for which license information can be found [here](https://www.industrydocuments.ucsf.edu/help/copyright/). ### Citation Information ```bibtex @inproceedings{harley2015icdar, title = {Evaluation of Deep Convolutional Nets for Document Image Classification and Retrieval}, author = {Adam W Harley and Alex Ufkes and Konstantinos G Derpanis}, booktitle = {International Conference on Document Analysis and Recognition ({ICDAR})}}, year = {2015} } ``` ### Contributions Thanks to [@dnaveenr](https://github.com/dnaveenr) for adding this dataset.
注释创建者: - 公开获取 语言来源: - 公开获取 语言: - 英语(en) 许可协议: - 其他 多语言属性: - 单语言 数据规模区间: - 10万 < 样本数 < 100万 源数据集: - 扩展自IIT-CDIP 任务类别: - 图像分类(image-classification) 任务子类别: - 多类别图像分类 PapersWithCode ID:rvl-cdip 数据集展示名称:RVL-CDIP 数据集查看器:未启用 数据集信息: 特征字段: - 名称:image 数据类型:图像 - 名称:label 数据类型: 分类标签: 类别名称: '0': 信件(letter) '1': 表单(form) '2': 电子邮件(email) '3': 手写文本(handwritten) '4': 广告(advertisement) '5': 科研报告(scientific report) '6': 学术出版物(scientific publication) '7': 规格说明书(specification) '8': 文件档案夹(file folder) '9': 新闻文章(news article) '10': 预算文档(budget) '11': 发票(invoice) '12': 演示文稿(presentation) '13': 调查问卷(questionnaire) '14': 简历(resume) '15': 备忘录(memo) 数据划分: - 名称:训练集 字节数:38816373360 样本数:320000 - 名称:测试集 字节数:4863300853 样本数:40000 - 名称:验证集 字节数:4868685208 样本数:40000 下载总大小:38779484559字节 数据集总大小:48548359421字节 # RVL-CDIP 数据集卡片 ## 目录 - [数据集描述](#dataset-description) - [数据集概述](#dataset-summary) - [支持任务与排行榜](#supported-tasks-and-leaderboards) - [语言](#languages) - [数据集结构](#dataset-structure) - [数据实例](#data-instances) - [数据字段](#data-fields) - [数据划分](#data-splits) - [数据集构建](#dataset-creation) - [构建初衷](#curation-rationale) - [源数据](#source-data) - [注释标注](#annotations) - [个人与敏感信息](#personal-and-sensitive-information) - [数据集使用注意事项](#considerations-for-using-the-data) - [数据集的社会影响](#social-impact-of-dataset) - [偏差分析](#discussion-of-biases) - [其他已知局限性](#other-known-limitations) - [附加信息](#additional-information) - [数据集维护者](#dataset-curators) - [许可协议信息](#licensing-information) - [引用信息](#citation-information) ## 数据集描述 - **主页:** [RVL-CDIP 数据集](https://www.cs.cmu.edu/~aharley/rvl-cdip/) - **代码仓库:** - **论文:** [深度卷积网络在文档图像分类与检索中的评估](https://arxiv.org/abs/1502.07058) - **排行榜:** [RVL-CDIP 数据集排行榜](https://paperswithcode.com/dataset/rvl-cdip) - **联系人:** [Adam W. Harley](mailto:aharley@cmu.edu) ### 数据集概述 RVL-CDIP(全称Ryerson Vision Lab Complex Document Information Processing,即瑞尔森视觉实验室复杂文档信息处理数据集)包含16个类别共40万张灰度图像,每个类别包含25000张图像。数据集划分为32万张训练图像、4万张验证图像与4万张测试图像。所有图像的最大维度均不超过1000像素。 ### 支持任务与排行榜 - `图像分类(image-classification)`:该任务的目标是将给定的文档图像分类为16个文档类型类别之一(如信件、表单等)。该任务的排行榜可参见[此处](https://paperswithcode.com/sota/document-image-classification-on-rvl-cdip)。 ### 语言 所有类别与文档均以英语作为主要语言。 ## 数据集结构 ### 数据实例 以下展示一个训练集的样本: { 'image': <Python图像处理库(PIL)Tiff图像插件.TiffImageFile image mode=L size=754x1000 at 0x7F9A5E92CA90>, 'label': 15 } ### 数据字段 - `image`:包含文档图像的`PIL.Image.Image`对象(Python图像处理库图像对象)。 - `label`:表示分类标签的整数。 <details> <summary>类别标签映射表</summary> json { "0": "信件(letter)", "1": "表单(form)", "2": "电子邮件(email)", "3": "手写文本(handwritten)", "4": "广告(advertisement)", "5": "科研报告(scientific report)", "6": "学术出版物(scientific publication)", "7": "规格说明书(specification)", "8": "文件档案夹(file folder)", "9": "新闻文章(news article)", "10": "预算文档(budget)", "11": "发票(invoice)", "12": "演示文稿(presentation)", "13": "调查问卷(questionnaire)", "14": "简历(resume)", "15": "备忘录(memo)" } </details> ### 数据划分 | |训练集|测试集|验证集| |----------|----:|----:|---------:| |样本数量|320000|40000|40000| 该数据集的划分比例参考了ImageNet的划分方式,其中32万张图像用于训练,4万张用于验证,剩余4万张用于测试。 ## 数据集构建 ### 构建初衷 摘自原论文: > 本工作公开了IIT-CDIP数据集的一个新的带标注子集,该子集包含16个类别共40万张文档图像,可用于训练用于文档分析任务的新型卷积神经网络(CNN,Convolutional Neural Network)。 ### 源数据 #### 初始数据收集与标准化处理 与IIT-CDIP数据集一致。 #### 源文本创作者 与IIT-CDIP数据集一致。 ### 注释标注 #### 标注流程 与IIT-CDIP数据集一致。 #### 标注人员 与IIT-CDIP数据集一致。 ### 个人与敏感信息 [需要更多信息] ## 数据集使用注意事项 ### 数据集的社会影响 [需要更多信息] ### 偏差分析 [需要更多信息] ### 其他已知局限性 [需要更多信息] ## 附加信息 ### 数据集维护者 该数据集由Adam W. Harley、Alex Ufkes与Konstantinos G. Derpanis三位作者维护。 ### 许可协议信息 RVL-CDIP是IIT-CDIP数据集的子集,其原始数据来自[Legacy烟草文档库](https://www.industrydocuments.ucsf.edu/tobacco/),该库的许可协议信息可参见[此处](https://www.industrydocuments.ucsf.edu/help/copyright/)。 ### 引用信息 bibtex @inproceedings{harley2015icdar, title = {深度卷积网络在文档图像分类与检索中的评估}, author = {Adam W Harley and Alex Ufkes and Konstantinos G Derpanis}, booktitle = {国际文档分析与识别会议(ICDAR,International Conference on Document Analysis and Recognition)}, year = {2015} } ### 贡献者 感谢[@dnaveenr](https://github.com/dnaveenr)添加该数据集。
数据集概述
数据集名称
- 名称: RVL-CDIP
- 别名: Ryerson Vision Lab Complex Document Information Processing
数据集基本信息
- 语言: 英语
- 许可证: 其他
- 多语言性: 单语
- 大小类别: 100K<n<1M
- 任务类别: 图像分类
- 任务ID: 多类图像分类
- 论文代码ID: rvl-cdip
数据集内容
- 图像数量: 400,000
- 类别数量: 16
- 训练集大小: 320,000图像
- 测试集大小: 40,000图像
- 验证集大小: 40,000图像
- 图像尺寸: 最大维度不超过1000像素
数据集结构
- 数据实例: 每个实例包含图像和标签
- 数据字段:
- image: 图像文件,类型为
PIL.Image.Image - label: 分类标签,类型为整数,对应16个类别
- image: 图像文件,类型为
- 数据分割:
- 训练集: 320,000图像
- 测试集: 40,000图像
- 验证集: 40,000图像
数据集来源
- 源数据集: 扩展自IIT-CDIP
数据集使用注意事项
- 许可证信息: 参考Legacy Tobacco Document Library
- 引用信息: bibtex @inproceedings{harley2015icdar, title = {Evaluation of Deep Convolutional Nets for Document Image Classification and Retrieval}, author = {Adam W Harley and Alex Ufkes and Konstantinos G Derpanis}, booktitle = {International Conference on Document Analysis and Recognition ({ICDAR})}}, year = {2015} }




