kavinh07/ocr_dataset_shamadhan_synth_5k_conj

Bengali Conjunct (যুক্তবর্ণ) OCR Dataset

2,996 synthetic images covering all ~295 Bengali consonant clusters. Each row includes the conjunct field so you can filter/oversample by cluster.

Split	Samples
train	2,996

Columns

Column	Type	Description
image	Image	Cropped word image (RGB)
text	string	Bengali word containing the conjunct
conjunct	string	Target conjunct cluster (e.g. ক্ষ, ন্ত্র)
class_name	string	conjunct_training
source	string	synthetic_conjunct

Load

from datasets import load_dataset
ds = load_dataset('parquet', data_files={'train': '/app/conjunct_output/hf/train/*.parquet'})