kavinh07/ocr_dataset_shamadhan_synth_30k

NID OCR Dataset

Split	Samples
train	28,347
validation	6,000

Sources

• Synthetic — TextRecognitionDataGenerator (token-balanced)
• Shamadhan — real cropped NID field images (reviewed labels)

Columns

Column	Type	Description
image	Image	Cropped NID field (RGB)
text	string	OCR ground-truth label
class_name	string	NID field type (e.g. en_name, address_line_00)
source	string	synthetic or shamadhan

Load

from datasets import load_dataset
ds = load_dataset('parquet', data_files={
    'train':      'hf_dataset/train/*.parquet',
    'validation': 'hf_dataset/validation/*.parquet',
})

Upload

ds.push_to_hub('your-org/nid-ocr-dataset')