Datasets:

MLap
/

Book-Scan-OCR

dataset_info:
  features:
    - name: filename
      dtype: string
    - name: text
      dtype: string
    - name: image
      dtype: image
  splits:
    - name: train
      num_bytes: 34202523
      num_examples: 156
  download_size: 32455168
  dataset_size: 34202523
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
language:
  - en
tags:
  - finetuning
  - VLM
  - OCR
  - Text
  - TextExtraction
size_categories:
  - n<1K
task_categories:
  - image-to-text

Best Usage

Suitable for fine-tuning Vision-Language Models (e.g., PaliGemma).

Dataset Creation

This dataset was generated using Mistral OCR and Google Lens, followed by manual cleaning for improved accuracy.

Image Source

Images are sourced from Sarvam.ai.

Please cite it as:

@dataset{Aman_MLap_BookScanOCR_2025,
  title        = {Book-Scan-OCR},
  author       = {Aman Prakash},
  year         = {2025},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/MLap/Book-Scan-OCR}
}