OmniDoc OCR Correction Bench: Document Images Paired with OCR-Text Prompts

A benchmark dataset pairs document images from OmniDocBench v1.5 with prompts containing PaddleOCR-extracted markdown text. The task is to correct OCR errors and restore proper formatting using the source image as reference. The dataset was created by author 'andynoodles' and was last updated on 2026-04-07.

Use Cases

• Benchmarking VLM performance on OCR error correction based on image-text pairs.
• Training models to restore document formatting based on the described source image reference.
• Evaluating document-to-markdown conversion accuracy using the described prompt structure.

Strengths

• Pairs document images with OCR-extracted text prompts, providing a direct benchmark for multimodal tasks.
• Specifically designed for evaluating VLMs on OCR correction and formatting, offering a clear task definition.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Pairs document images from OmniDocBench v1.5 with PaddleOCR-extracted markdown text.

Freshness

Last updated 2026-04-07 10:41:24; freshness should be verified.