Split
Samples
train
145,800
validation
35,363
Synthetic — TextRecognitionDataGenerator (token-balanced)
Shamadhan — real cropped NID field images (reviewed labels)
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… See the full description on the dataset page: https://huggingface.co/datasets/kavinh07/ocr_dataset_shamadhan_synth_30k_p2. 
