stefan-hf/rvlcdip-o

RVL-CDIP-O

RVL-CDIP-O is an out-of-distribution test set for document classifiers trained on RVL-CDIP. It contains 2,915 document page images that belong to none of the sixteen RVL-CDIP categories: court filings, instructional guides, sheet music, timelines, schedules, and other document types that the benchmark's label set does not cover. A classifier trained on RVL-CDIP should recognize that these pages are out of scope instead of assigning one of its sixteen labels with high confidence.

The set was introduced, together with RVL-CDIP-N, in Evaluating Out-of-Distribution Performance on Document Image Classifiers (NeurIPS 2022 Datasets and Benchmarks).

The set has been publicly available since that paper was published; the original release is at github.com/gxlarson/rvl-cdip-ood. This repository contains the same pages as JPEG files together with their OCR.

Contents

• 2,915 JPEG images in a single test split. There are no labels: every page is out of scope.
• A text field with the OCR text of each page (Amazon Textract, one line of text per detected line; empty for six pages on which no text was found).
• The raw Textract responses, with word and line bounding boxes, in textract/, one JSON file per page with the same file stem as the image.
• Pages are in color and 1,000 pixels tall, converted from the TIFF files of the original release.
• The documents were collected from DocumentCloud and from web search, and are largely born-digital and more recent than the tobacco-litigation scans in RVL-CDIP.

Use

The usual protocol scores each page with a confidence measure from a trained classifier, such as the maximum softmax probability or the energy score, and reports how well that score separates RVL-CDIP-O from in-domain pages (the RVL-CDIP test split, or RVL-CDIP-N).

RVL-CDIP is grayscale and these images are in color. Convert them to grayscale before scoring, as RVL-CDIP-trained models expect; otherwise color alone marks the pages as out of distribution.

from datasets import load_dataset

ds = load_dataset("stefan-hf/rvlcdip-o", split="test")
image = ds[0]["image"].convert("L")

Related resources

• Classifiers trained on RVL-CDIP: stefan-hf/rvlcdip-classifiers
• Identification codes in RVL-CDIP: stefan-hf/rvlcdip-id-codes

License

The selection of the documents is released under CC BY 4.0. The documents themselves were collected from DocumentCloud and from web search, and they and the OCR text derived from them remain the property of their original authors and publishers; they are redistributed here for research on document classification. The pages have not been reviewed for personal data. If you hold the rights to a document, or find a page that should not be public, please open a discussion on the Community tab of this dataset and the page will be removed.

Citation

@inproceedings{larson-etal-2022-rvlcdipn,
  author    = {Larson, Stefan and Lim, Yi Yang Gordon and Ai, Yutong and Kuang, David and Leach, Kevin},
  title     = {Evaluating Out-of-Distribution Performance on Document Image Classifiers},
  booktitle = {Advances in Neural Information Processing Systems},
  volume    = {35},
  pages     = {11673--11685},
  publisher = {Curran Associates, Inc.},
  year      = {2022},
  url       = {https://proceedings.neurips.cc/paper_files/paper/2022/file/4c0986bd04d747745beba3752bdf4d9d-Paper-Datasets_and_Benchmarks.pdf}
}