Line-level text recognition dataset extracted from Moroccan Bulletin Officiel (BO) PDF documents.
Built for fine-tuning PP-OCRv5 Arabic recognition on French/Arabic administrative text.
Metric
Value
Total samples
12649
Train
9996
Test
2653
Original train
1666
Augmented train
8330
Unique charset
141
Arabic ratio
1.24%
Label length p50/p90/p99
35.0/40.0/40.0