RVL-CDIP Document Images with 16 Classes

The RVL-CDIP dataset contains 400,000 grayscale images of documents, evenly distributed across 16 classes with 25,000 images per class. It is split into 320,000 training, 40,000 validation, and 40,000 test images. The dataset was created by aharley for document information processing tasks.

Use Cases

• Train a multi-class image classification model to predict the 16 document classes from grayscale images.
• Benchmark document classification performance using the predefined 320,000-image training set and 40,000-image validation set.
• Analyze the distribution of document types across the 16 classes, each containing 25,000 samples.
• Develop a grayscale image feature extractor for complex document information processing tasks.

Strengths

• Large scale of 400,000 total images provides substantial data for model training.
• Balanced class distribution with exactly 25,000 images in each of the 16 classes.
• Predefined splits of 320,000 training, 40,000 validation, and 40,000 test images facilitate standardized evaluation.

Limitations

• Images are grayscale, lacking color information that may be present in original documents.
• Geographic or source bias is possible as the dataset is noted with a 'Region:us' tag.
• The dataset's specific document sources and collection methodology are not detailed in the provided input.

Provenance

Source

Ryerson Vision Lab (RVL), extended from the IIT-CDIP dataset.

Collection Method

Collection and processing of complex document images, as indicated by the dataset name and tags.

Freshness

The dataset was last updated on the platform in September 2024.

Geography

Primarily United States, as suggested by the 'Region:us' tag.