Downloads

🤗 HuggingFace | 🔗 GitHub | 🦢 Kaggle | 📃Paper

PsOCR Dataset

PsOCR is a large-scale synthetic Optical Character Recognition (OCR) dataset developed for the low-resource Pashto language. It is the first comprehensive publicly available Pashto OCR dataset, containing one million synthetic images with annotations at word, line, and document levels. The dataset covers extensive variations in typography, including 1,000 unique font families, diverse colors, image resolutions, and text layouts. PsOCR also introduces the first standardized Pashto OCR benchmark with 10K test images for systematic evaluation and comparison of OCR models. The dataset aims to advance research in Pashto OCR and promote the development of robust AI systems for low-resource languages

Dataset Usage Policy

The PsOCR dataset is strictly prohibited for commercial use. It is provided exclusively for educational and research purposes. Any use of the dataset in commercial products, services, applications, or profit-generating activities is not permitted without prior written permission.

Publicly Available Test Set

The PsOCR test set (benchmark) containing 10K images is publicly available on Hugging Face and can be downloaded freely without any restrictions. This benchmark is provided to support fair evaluation and comparison of OCR systems for the low-resource Pashto language.

Access to 100K Image Subset

A subset containing 100K images can be provided to researchers for academic and research purposes.

To request access:

  1. Download the Dataset Request Form.
  2. Complete the form with the required information.
  3. The form must be signed by an academic researcher with a minimum academic rank of Assistant Professor (or equivalent).
  4. Send the completed form to us.

After reviewing your request, we will provide:

  • A download link to the dataset .zip file.
  • The password required to extract the archived dataset files.

Access to Full 1M Image Dataset

The complete daaset of 1 million images is also available upon request. Access to the full dataset requires a licensing fee of USD $980.

Accepted Payment Methods
  • EasyPaisa
  • WeChat
  • AliPay
  • Bank Wire Transfer
  • USDT Cryptocurrency via Binance, Bitget, or OKX

For dataset access requests or further inquiries, please contact us at contact@zirak.ai