fgaim/GLOCR-Tigrinya

---
license: cc-by-4.0
task_categories:
- image-to-text
language:
- ti
tags:
- ocr
- tigrinya
- geez-script
- text-recognition
- geezlab
size_categories:
- 100K<n<1M
pretty_name: GLOCR - GeezLab OCR Dataset
configs:
- config_name: news
data_files:
- split: train
path: data/news/train.parquet
- split: validation
path: data/news/validation.parquet
- split: test
path: data/news/test.parquet
- config_name: bible
data_files:
- split: train
path: data/bible/train.parquet
- split: validation
path: data/bible/validation.parquet
- split: test
path: data/bible/test.parquet
- config_name: top150k
data_files:
- split: train
path: data/top150k/train.parquet
- split: validation
path: data/top150k/validation.parquet
- split: test
path: data/top150k/test.parquet
- config_name: characters
data_files:
- split: train
path: data/characters/train.parquet
- split: validation
path: data/characters/validation.parquet
- split: test
path: data/characters/test.parquet
- config_name: unsegmented
data_files:
- split: train
path: data/unsegmented/train.parquet
- config_name: all
data_files:
- split: train
path: data/*/train.parquet
- split: validation
path: data/*/validation.parquet
- split: test
path: data/*/test.parquet
default: true
---

# GLOCR: GeezLab OCR Dataset

## Overview

GLOCR is a Text Recognition (TR) and Optical Character Recognition (OCR) dataset for the **Tigrinya language**. The dataset contains a total of 661K image-label pairs from multiple data sources. In addition to the characters-only data, the major part of the dataset is a collection of multi-word text images with labels from three categories: News (from Haddas Ertra newspaper), the Bible, and random-trigrams of the 150k most common words in Tigrinya.

### Dataset Summary

- **Total samples**: ~661K image-label pairs
- **Total size**: >1.3GB (tar.gz archives)
- **DOI**: [10.7910/DVN/RQTSD2](https://doi.org/10.7910/DVN/RQTSD2)

### Dataset Subsets

| Config | Description | Train | Validation | Test |
|--------|-------------|------:|-----------:|-----:|
| `news` | Newspaper text-lines | 200K | 15K | 15K |
| `bible` | Biblical text-lines | 80K | 10K | 10K |
| `top150k` | Word trigrams | 150K | 15K | 15K |
| `characters` | Single characters | 120K | 15K | 15K |
| `unsegmented` | Full-page scans | 506 | - | - |

## Usage

### Loading a specific subset

```python
from datasets import load_dataset

# Load a specific subset, one of: news, bible, top150k, characters, unsegmented
news = load_dataset("fgaim/GLOCR-Tigrinya", "news")

# Access samples
sample = news["train"][0]
print(sample["text"])
sample["image"].show()
```

### Loading a specific split

```python
# Load a specific split of a subset
bible_test = load_dataset("fgaim/GLOCR-Tigrinya", "bible", split="test")

# Access samples
print(bible_test["text"][0])
bible_test["image"][0].show()
```

### Loading all text-line data combined

```python
# Load all text-line data combined
all_data = load_dataset("fgaim/GLOCR-Tigrinya", "all")

# Access samples
sample = all_data["train"][0]
print(sample["text"])
sample["image"].show()
```

## Links

- [Harvard Dataverse](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/RQTSD2)
- [GitHub Repository](https://github.com/fgaim/GLOCR)

## Citation

```bibtex
@data{gaim-2021-glocr,
title = {{GLOCR: GeezLab OCR Dataset}},
author = {Fitsum Gaim},
year = {2021},
month = {April},
publisher = {Harvard Dataverse},
version = {1.0},
doi = {10.7910/DVN/RQTSD2},
url = {https://doi.org/10.7910/DVN/RQTSD2},
dataverse = {https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/RQTSD2}
}
```

## License

This work is licensed under a [Creative Commons Attribution-ShareAlike 4.0 International License](https://creativecommons.org/licenses/by-sa/4.0/).

<a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/"><img alt="Creative Commons License" style="border-width:0" src="https://licensebuttons.net/l/by-sa/4.0/88x31.png" /></a>