Mahmudisnan/Tulak-tulak

Description

Tulak-tulak is a traditional writing system associated with the Mandailing people of North Sumatra, Indonesia. Mandailing script is considered a distinct writing system with its own characteristic forms that distinguish it from other traditional scripts in the region. However, its use has become increasingly limited, particularly in written narratives, highlighting the need for documentation and digital preservation. This paper introduces a comprehensive, age-diverse handwritten Mandailing script dataset collected from 184 respondents, primarily representing students from different educational levels. Each respondent was asked to write each of the 21-character classes exactly five times. Following data collection and preprocessing, the dataset comprised 19,320 original images, with 920 images per class, stored in JPG format at a resolution of 512 × 512 pixels. To support model development and evaluation, the dataset was partitioned into three subsets, with approximately 70% allocated to training, 20% to validation, and 10% to testing. The training subset contains 13,545 original and 460,530 augmented images, the validation subset contains 3,885 original and 132,090 augmented images, and the testing subset contains 1,890 original and 64,260 augmented images. The dataset can be accessed on Hugging Face under a CC BY 4.0 license and is intended for OCR model development and Mandailing script digitization. The dataset addresses a significant gap in existing benchmarks, which is expected to support future research and real-world educational applications by facilitating the development of robust and accurate recognition models for Mandailing script.

Repository Structure

Tulak-tulak/
├── Dataset/
│   ├── Dataset_image_metadata.csv
│   ├── Original/                         # 19,320 original images (approx. 845 MB)
│   │   ├── split_manifest.csv
│   │   ├── train/                        # 13,545 images
│   │   │   ├── Char 01/
│   │   │   │   └── Char_01_<id>.jpg
│   │   │   ├── Char 02/
│   │   │   ├── ...
│   │   │   └── Char 21/
│   │   ├── val/                          # 3,885 images
│   │   │   ├── Char 01/
│   │   │   ├── ...
│   │   │   └── Char 21/
│   │   └── test/                         # 1,890 images
│   │       ├── Char 01/
│   │       ├── ...
│   │       └── Char 21/
│   │
│   └── augmentation/                     # 656,880 augmented images (approx. 30 GB)
│       ├── train/                        # 460,530 images
│       │   ├── Char 01/
│       │   │   └── Char_01_<id>__aug_<number>_<technique>.jpg
│       │   ├── Char 02/
│       │   ├── ...
│       │   └── Char 21/
│       ├── val/                          # 132,090 images
│       │   ├── Char 01/
│       │   ├── ...
│       │   └── Char 21/
│       └── test/                         # 64,260 images
│           ├── Char 01/
│           ├── ...
│           └── Char 21/
│
├── Uncropped/                            # 1,840 uncropped images (approx. 5.6 GB)
│   ├── 1_1.jpg
│   ├── 1_2.jpg
│   ├── ...                               # Group 1: 920 images
│   ├── 1_920.jpg
│   ├── 2_1.jpg
│   ├── 2_2.jpg
│   ├── ...                               # Group 2: 920 images
│   └── 2_920.jpg
│
├── Dataset.zip                         # Dataset archive (approx. 29 GB)
├── Uncropped.zip                       # Uncropped archive (approx. 3.6 GB)
├── Tulak_tulak_form.pdf
└── aug_tulak_tulak.ipynb

Image count summary

Data group	Train	Validation	Test	Total
Original	13,545	3,885	1,890	19,320
Augmentation	460,530	132,090	64,260	656,880
Uncropped	-	-	-	1,840
Grand total				678,040

Data overview

• The classified dataset contains 21 classes, from Char 01 through Char 21.
• The classified data is divided into train, val, and test splits.
• Original images follow the Char_<class>_<id>.jpg filename pattern.
• Augmented images follow the Char_<class>_<id>__aug_<number>_<technique>.jpg filename pattern.
• Each original image has 34 augmented variants.
• The Augmentation directory contains augmented images only; copies of the original images have been removed.
• The Uncropped directory has not been divided into splits or class directories.
• The Tulak_tulak.pdf is the form that is used to collect the dataset
• The aug_tulak_tulak.ipynb provides code to process augmentation for all original images