PubTables V2: Large-Scale Full-Page and Multi-Page Table Extraction

PubTables-v2 is a dataset for full-page and multi-page table extraction, created by kensho and released in February 2026. It is described as large-scale and comes in 3 collections, each containing tables in a specific context.

Use Cases

• Train computer vision models for full-page table extraction from document images.
• Develop multi-page table extraction algorithms using the dataset's specific context collections.
• Benchmark table parsing performance on a large-scale dataset tagged for document analysis.

Strengths

• Described as a 'large-scale' dataset for table extraction.
• Contains 3 distinct collections of tables, each in a specific context.
• Specifically designed for both full-page and multi-page table extraction tasks.

Limitations

• The exact size, row count, column structure, and sample data are unknown.
• Specific file formats and licensing terms are not provided in the input.
• The dataset's composition and potential biases within its 3 collections are unspecified.

Provenance

Source

huggingface

Freshness

Last updated on 2026-03-19.