BanglaNet-45: Handwritten Bangali Character Dataset

BanglaNet-45: Handwritten Bengali Character Dataset

Description:

A comprehensive handwritten Bengali character dataset
comprising 45 character classes including vowels and
consonants. Collected from multiple writers and
preprocessed at 300 DPI.

Dataset Statistics:

- Total Classes : 45
- Real Images : 4487
- Aug Images : 13800
- Image Size : 32x32 pixels
- Color : Grayscale (Inverted)
Black BG + White Text
- Format : PNG
- Train/Val/Test : 70% / 15% / 15%

Character Classes:

Vowels (11): অ আ ই ঈ উ ঊ ঋ এ ঐ ও ঔ
Consonants (34): ক খ গ ঘ ঙ চ ছ জ ঝ ঞ
ট ঠ ড ড় ঢ ঢ় ণ ত থ দ ধ
ন প ফ ব ভ ম য র ল শ ষ স হ য়

Folder Structure:

real_data/
train/ - 70% of real images per class
val/ - 15% of real images per class
test/ - 15% of real images per class

augmented_train/
- Real train + Augmented images
- Target: 300 images per class

Augmentation Techniques:

- Rotation : ±12 degrees
- Affine : scale(0.88-1.12),
translate(±6%), shear(±8°)
- Elastic : alpha=1.5, sigma=8
- Brightness : ±15%
- Gaussian Noise : var(3-15)
- Grid Distortion: limit=0.15

CSV Files:

- full_dataset.csv : Complete dataset info
- train.csv : Training set
- val.csv : Validation set
- test.csv : Test set
- label_map.csv : Class label mapping

License:

CC BY 4.0 - Free to use with attribution

Citation:

[Nasir, Md Nasir Uddin; Sultana, Masrifa; Pranto, Md. Shoriful Islam (2026), “BanglaNet-45: Handwritten Bangali Character Dataset”, Mendeley Data, V1, doi: 10.17632/47fpd8js36.1]