This dataset contains segmented line-level images and corresponding transcriptions in Luganda, a low-resource Bantu language spoken primarily in Uganda. It was created to support research in optical character recognition (OCR) and handwritten/printed text recognition for under-resourced African languages.

The dataset was developed as part of an OCR research project focused on building end-to-end deep learning models for text recognition in low-resource settings. It is suitable for training and evaluating modern OCR architectures such as CRNN, TrOCR, and VisionEncoderDecoder models.

Each sample consists of:

an image of a single text line
its ground-truth transcription
the length of the transcription (number of characters)

Supported Tasks

Optical Character Recognition (OCR)
Handwritten Text Recognition (HTR)
Low-resource language modeling
Vision-to-text sequence modeling

Languages

Luganda (lg)

Dataset Structure

Each example contains the following fields:

image: An image of a text line (PNG format)
text: The corresponding Luganda transcription
transcript_length: Length of the transcription in characters

Example:

image	text	transcript_length
🖼️	"Amaviivi gamulumira munda era gazimbye"	38

Dataset Splits

Split	Samples
Train	17,586

(Note: Users are encouraged to create their own validation/test splits depending on experimental needs.)

Intended Use

This dataset is intended for:

Training OCR models for Luganda
Benchmarking low-resource OCR performance
Fine-tuning multilingual OCR systems
Academic research in document analysis and African language technologies

Limitations

The dataset size is relatively small compared to high-resource OCR datasets.
It may contain noise due to segmentation or transcription errors.
The data may not cover all Luganda writing styles or fonts.

Models trained on this dataset may not generalize to:

highly cursive handwriting
historical manuscripts
heavily degraded documents

Citation

If you use this dataset in your work, please cite the associated paper:

Downloads last month: 7