PubTables-1M: One Million Tables Extracted from Unstructured Documents

PubTables-1M is a large-scale dataset for training models to detect and understand tables in unstructured documents. The dataset, created by Microsoft Research, is associated with the Table Transformer model and was last updated on Hugging Face in August 2023. It is designed to advance comprehensive table extraction from scientific and other document types.

Use Cases

• Training table detection models based on the provided detection model mentioned in the description
• Developing table structure recognition systems based on the structure recognition model mentioned in the description
• Benchmarking performance of document layout analysis tools on a large-scale corpus
• Improving information extraction pipelines for scientific and technical documents

Strengths

• Large-scale dataset with one million tables, as indicated by the title
• Created by Microsoft Research, a major institution in the field
• Associated with a published paper and the open-source Table Transformer model

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Last updated 2023-08-08 16:43:14; freshness should be verified

Provenance

Source

Microsoft Research (bsmock on Hugging Face)

Collection Method

Extracted from unstructured documents, likely scientific publications.

Time Range

null

Freshness

Last updated 2023-08-08 16:43:14

Geography

null