Datasets:
Tasks:
Image-to-Text
Formats:
parquet
Languages:
Arabic
Size:
10K - 100K
Tags:
optical-character-recognition
handwritten-text-recognition
historical-documents
arabic
naskh
taliq
License:
Link ICDAR 2026 paper and MAS GitHub repository
Browse files
README.md
CHANGED
|
@@ -24,6 +24,13 @@ MAS is a line-level OCR benchmark containing 11,841 expert-annotated lines
|
|
| 24 |
from nine authentic Arabic manuscript works. It covers Naskh, Taliq, and
|
| 25 |
Nastaliq scripts and manuscripts copied between the 12th and 20th centuries.
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
## Dataset structure
|
| 28 |
|
| 29 |
The dataset exposes one configuration, `mas`, with two splits:
|
|
|
|
| 24 |
from nine authentic Arabic manuscript works. It covers Naskh, Taliq, and
|
| 25 |
Nastaliq scripts and manuscripts copied between the 12th and 20th centuries.
|
| 26 |
|
| 27 |
+
This dataset accompanies the ICDAR 2026 paper *A Millennium of Arabic
|
| 28 |
+
Manuscripts in Three Styles: A Line-Level OCR Benchmark for Naskh, Taliq,
|
| 29 |
+
and Nastaliq*.
|
| 30 |
+
|
| 31 |
+
Source code, training configurations, and evaluation utilities are available
|
| 32 |
+
in the [MAS GitHub repository](https://github.com/ai-forever/MAS).
|
| 33 |
+
|
| 34 |
## Dataset structure
|
| 35 |
|
| 36 |
The dataset exposes one configuration, `mas`, with two splits:
|