CIVQA TesseractOCR LayoutLM: Czech Invoice Visual Question Answering Dataset

CIVQA TesseractOCR LayoutLM is a dataset for Visual Question Answering on Czech invoices, created by fimu-docproc-research. The dataset was generated using Tesseract OCR and is pre-encoded for the LayoutLM model, focusing on 15 key invoice entities. It was last updated on November 21, 2023.

Use Cases

• Fine-tuning LayoutLM for invoice information extraction based on the 15 annotated entities.
• Training Visual Question Answering models on structured document data.
• Benchmarking OCR and document understanding pipelines for Czech financial documents.
• Researching multimodal learning approaches that combine text and layout information from invoices.

Strengths

• Focuses on 15 different entities crucial for invoice processing.
• Invoices were obtained from public sources, suggesting a basis in real-world documents.
• Dataset is pre-encoded for the LayoutLM model, potentially reducing preprocessing effort.

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.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

fimu-docproc-research on Hugging Face.

Collection Method

Created with Tesseract OCR from invoices obtained from public sources.

Freshness

Last updated 2023-11-21 20:45:53; freshness should be verified.

Geography

Likely Czech Republic, based on the language of the invoices.