Om-7387/Marathi-OCR

--- license: apache-2.0 task_categories: - image-to-text - translation - table-to-text language: - mr - en - hi tags: - legal - agent pretty_name: Marathi OCR model training dataset size_categories: - n<1K --- # Marathi OCR Model Training Dataset ## Description This dataset is designed for training Optical Character Recognition (OCR) models for Marathi text. It contains paired image and text data where each image includes Marathi textual content, and the corresponding ground truth text is provided. ## Dataset Structure The dataset consists of: - **PNG Images Folder** - Contains input images with Marathi text. - **CSV File** - Columns: - `_id`: Unique identifier - `image_path`: Relative path to the image - `text`: Ground truth Marathi text Example: | _id | image_path | text | |-----|------------------|------------------| | 1 | images/img1.png | नमस्कार | | 2 | images/img2.png | माझं नाव ओमकार आहे | ## Use Cases - OCR model training (TrOCR, CRNN, Vision Transformers) - Marathi document digitization - Text recognition in low-resource languages ## Data Format - Image format: `.png` - Annotation format: `.csv` - Encoding: UTF-8 (supports Marathi text) ## Dataset Size - Total samples: ~5,000 ## Notes - Ensure image paths in CSV are correctly mapped. - Text is normalized and aligned with images. - Suitable for supervised learning tasks. ## Loading Dataset ```python from datasets import load_dataset dataset = load_dataset("Om-7387/Marathi-OCR") ---