| --- |
| language: |
| - bn |
| task_categories: |
| - image-to-text |
| tags: |
| - ocr |
| - bangla |
| - document-understanding |
| - vision-language |
| --- |
| |
|
|
| # Bangla OCR Validation Dataset (Printed + Scanned) |
|
|
| ## 📌 Description |
|
|
| This dataset is a Bangla OCR validation dataset containing a mix of printed document images and their corresponding text annotations. It is designed to evaluate OCR and vision-language models on both clean digital text and scanned document images. |
|
|
| ## 📊 Dataset Composition |
|
|
| - 1507 **line-level images** with text annotations |
| - 50 **full-page document images** with text |
| - Data includes: |
| - Printed/typed Bangla text (clean) |
| - Scanned document images (noisy, real-world) |
|
|
| ## 🧾 Features |
|
|
| Each sample contains: |
|
|
| - `image`: Input image (line or page) |
| - `text`: Ground-truth Bangla transcription |
| - `type`: Indicates data type (`line` or `page`) |
|
|
| ## 🎯 Use Cases |
|
|
| - Bangla OCR evaluation |
| - Document understanding |
| - Vision-language model validation |
| - Robustness testing (clean vs scanned) |
|
|
| ## 🚀 Usage |
|
|
| ```python |
| from datasets import load_dataset |
| |
| dataset = load_dataset("arobin79/bangla-ocr-validation_data_printed") |
| |
| sample = dataset["train"][0] |
| print(sample["text"]) |
| sample["image"].show() |
| |
