RVL-CDIP Document Image Classification Subset

RVL-CDIP 10 Examples Per Class is a curated subset of the RVL-CDIP dataset, hosted by author nielsr on Hugging Face. It provides a smaller sample for tasks like document image classification, with 10 examples per class. The dataset was last updated in August 2022.

Use Cases

• Train a convolutional neural network to classify document types using image data.
• Benchmark model performance on a subset of the RVL-CDIP document categories.
• Test data augmentation techniques on labeled document images to improve generalization.

Strengths

• Derived from the established RVL-CDIP dataset, which contains over 400,000 grayscale images.
• Provides a balanced sample with 10 examples per class, suitable for quick prototyping.

Limitations

• The sample size is very small, limiting its use for training anything but the simplest models.
• Contains only a tiny fraction of the full RVL-CDIP dataset's scope and diversity.

Provenance

Source

Subset of the RVL-CDIP (Ryerson Vision Lab Complex Document Information Processing) dataset.

Collection Method

Sampled from the larger collection of scanned document images.

Freshness

Last updated in 2022.

Geography

Region tags suggest a US focus, but specific coverage is unknown.