ocr-bench-moh / README.md
davanstrien's picture
davanstrien HF Staff
Add PP-OCRv6-medium OCR results (50 samples) [pp-ocrv6]
770da8d verified
|
Raw
History Blame
2.25 kB
metadata
tags:
  - ocr
  - text-recognition
  - tesseract
  - uv-script
  - generated
dataset_info:
  config_name: pp-ocrv6
  features:
    - name: image
      dtype: image
    - name: b_number
      dtype: string
    - name: page_index
      dtype: int64
    - name: source_row
      dtype: int64
    - name: markdown
      dtype: string
    - name: pp_ocr_blocks
      dtype: string
    - name: inference_info
      dtype: string
  splits:
    - name: train
      num_bytes: 20968020
      num_examples: 50
  download_size: 20577145
  dataset_size: 20968020
configs:
  - config_name: pp-ocrv6
    data_files:
      - split: train
        path: pp-ocrv6/train-*

Document OCR using Tesseract

This dataset contains OCR results from images in davanstrien/moh-bench-sample using Tesseract, the classical open-source CPU OCR engine — a cheap, no-GPU baseline alongside the VLM OCR recipes.

Processing Details

Configuration

  • Image Column: image
  • Output Column: markdown
  • Dataset Split: train
  • Page Segmentation Mode (psm): 3
  • OCR Engine Mode (oem): 3
  • Workers: 64

Model Information

Tesseract is a classical (non-VLM) OCR engine:

  • Runs on CPU — no GPU required
  • v4+ uses an LSTM-based recognition engine
  • 100+ languages via installable data packs
  • Plain-text output (no markdown / table / formula structure)
  • Apache-2.0 licensed

Dataset Structure

The dataset contains all original columns plus:

  • markdown: The recognised text (plain text)
  • inference_info: JSON list tracking all OCR models applied to this dataset

Reproduction

uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/tesseract-ocr.py \
    davanstrien/moh-bench-sample \
    <output-dataset> \
    --image-column image \
    --lang eng \
    --psm 3

Generated with UV Scripts