Scanned Images Dataset for OCR and VLM finetuning

This dataset is a meticulously curated collection of scanned document images across 10 diverse categories, designed to advance Optical Character Recognition (OCR) technology and optimize Vision-Language Models (VLMs). With its rich variety of real-world document types and textual layouts, this dataset is an indispensable resource for researchers and developers aiming to enhance AI-powered document processing.

Dataset Categories and Key Features

1. Advertisements

   • Includes promotional materials with a mix of text, graphics, and unique layouts.
   • Perfect for testing OCR systems on mixed-media documents.

2. Emails

   • Features scanned correspondence with structured headers, signatures, and body content.
   • Useful for tasks like information extraction and text recognition.

3. Forms

   • Contains predefined templates such as application forms and invoices with structured data fields.
   • Ideal for training OCR systems to recognize tabular and structured data.

4. Letters

   • Offers formal and informal correspondence in handwritten and printed formats.
   • Perfect for studying text style variations.

5. Memos

   • Focuses on internal business communications with concise text and structured headers.
   • Provides variability for OCR layout training.

6. News Articles

   • Comprises scanned clippings with columns, headlines, and images.
   • Challenges OCR systems with multi-column text and visual elements.

7. Notes

   • Includes informal handwritten or typed notes with irregular formatting.
   • Simulates real-world, unstructured text challenges.

8. Reports

   • Features documents with structured sections, tables, and graphs like financial or project reports.
   • Enhances OCR performance on professional documents.

9. Resumes

   • Contains personal profiles with structured fields, tabular data, and creative layouts.
   • Supports information extraction and classification tasks.

10. Scientific Papers

    • Includes dense text, equations, tables, and diagrams.
    • Tailored for academic OCR and VLM fine-tuning.