Invoice Document Images from RVL-CDIP Subset

A subset of the RVL-CDIP collection containing only invoice images. The full RVL-CDIP dataset consists of 400,000 grayscale images across 16 classes, with 25,000 images per class. It includes 320,000 training images, 40,000 validation images, and 40,000 test images.

Use Cases

• Train a document classifier to identify invoice images within the 16 RVL-CDIP document classes.
• Fine-tune a vision model on grayscale invoice images where the largest dimension does not exceed 1000 pixels.
• Use the structured 320k/40k/40k train/validation/test split for model evaluation on invoice documents.

Strengths

• Contains 400,000 grayscale images from the well-known RVL-CDIP benchmark.
• Provides a balanced distribution with 25,000 images per class across 16 document types.
• Includes predefined splits of 320,000 training, 40,000 validation, and 40,000 test images.

Limitations

• This is only a subpart of the original dataset, limiting scope to invoice images.
• Specific column metadata and file formats for the subset are not provided in the input.
• The dataset's last update was in April 2022, which may not reflect current document formats.

Provenance

Source

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

Collection Method

Subset extracted from the original RVL-CDIP image collection.

Freshness

2022-04-06

Geography

Region tag suggests 'us', but specific coverage is unknown.