PALM-LAY: A Multi-Script Cross-Regional Dataset for Layout Analysis of Palm Leaf Manuscripts

---

📖 Overview

PALM-LAY is the first unified, cross-regional annotated dataset specifically designed for layout analysis of historical palm leaf manuscripts.

Palm leaf manuscripts are among Asia’s oldest written heritage, covering religious scriptures, literary epics, astrology charts, and indigenous medical treatises. Their script diversity, complex layouts, and physical degradation pose unique challenges for document image analysis.

PALM-LAY addresses this gap by providing:

• 566 manuscript pages curated from six distinct collections.
• 6,334 annotated layout regions across seven categories:
  • MainRegion
  • ParagraphRegion
  • TextLineRegion
  • SymbolicMark
  • PhysicalDamage
  • Illustration
  • Other

The dataset enables benchmarking of state-of-the-art object detection models for cross-script generalization and multi-class layout segmentation.

---

🌍 Scripts & Sources

The dataset spans six palm leaf traditions from South and Southeast Asia:

• Tamil (India) – Naladiyar, Tholkapppiyam, Thirikadugam
• Jathakam (India, Malayalam) – astrological horoscopes
• Kambaramayanam (India) – Tamil epic manuscripts
• Khmer (Cambodia) – Sleukrith and archive collections
• Sundanese (Indonesia, Java) – 15th century manuscripts
• Balinese (Indonesia, Bali) – Balinese Lontar collections

Fig. 1 – Examples of palm leaf manuscripts across six scripts and regions.

---

🗂️ Layout Annotation Schema

Each image is annotated using seven consistent region categories:

Category	Description
MainRegion	Primary content area; whole text blocks
ParagraphRegion	Grouped blocks of related text lines
TextLineRegion	Individual horizontal lines of text
SymbolicMark	Section dividers, religious symbols
PhysicalDamage	Binding holes, cracks, fading
Illustration	Deities, animals, cultural drawing
Other	Non-original (stamps, labels, notes)

Fig. 2 – Visualization of annotated regions across scripts.

---

📊 Dataset Statistics

Script	Pages	Train	Test
Tamil	101	81	20
Jathakam	108	86	22
Kambaramayanam	41	33	8
Khmer	155	124	31
Balinese	100	80	20
Sundanese	61	49	12
Total	566	453	113

Fig. 3 – Annotation workflow and quality control pipeline.

---

⚙️ Benchmarking Experiments

Experiments:

1. Script-specific performance – trained & tested per script.
2. Cross-script generalization – trained on combined dataset, tested across all scripts.

Key Findings

• High accuracy on MainRegion, ParagraphRegion, and TextLineRegion.
• Lower accuracy on small/rare categories (SymbolicMark, PhysicalDamage, Other).
• YOLO series excelled on small-object detection.
• Transformer-based DETR models handled large, structured regions well.
• Cross-script training improved underrepresented categories, showing feature transfer across scripts.

Fig. 4 – Sample detection outputs on different scripts.

Quick Sample Run

pip install -r requirements.txt

# YOLOv8 / YOLOv9 / YOLOv11
python main.py train-yolo* --weights yolov**.pt --epochs 100

# DETR
python main.py train-detr --epochs 100 --batch 4

# RF-DETR (external repo required)
python main.py train-rfdetr \
  CONFIG=configs/rf_detr.py

🙏 Acknowledgements

• Manuscript collections sourced from India, Cambodia and Indonesia under open Creative Commons license.

• The annotation team included students and researchers from Cambodia, China, and Indonesia.

• Funding Support: This is part of the PALM-WORLD project is supported by The World Academy of Sciences (Italy), the Chinese Academy of Sciences (China), One-to-Many Research (Cambodia), and the National Natural Science Foundation of China (China).`

• Leading Project By: Nimol Thuon

🔗 References

• Nair, B.B., Rani, N.S. (2023). *HMPLMD: Handwritten Malayalam palm leaf manuscript dataset.
• Jailingeswari, I., Gopinathan, S. (2024). *Tamil handwritten palm leaf manuscript dataset (TH-PLMD).
• Valy, D., Verleysen, M., Chhun, S., Burie, J.C. (2017). *A new Khmer palm leaf manuscript dataset for document analysis and recognition: Sleukrith Set.
• Suryani, M., Paulus, E., Hadi, S., Darsa, U.A., Burie, J.C. (2017). The handwritten Sundanese palm leaf manuscript dataset from 15th century.
• Kesiman, M.W.A., Burie, J.C., Wibawantara, G.N.M.A., Sunarya, I.M.G., Ogier, J.M. (2016). *AMADI_LontarSet: The first handwritten Balinese palm leaf manuscripts dataset.
• Kesiman, M.W.A., Valy, D., Burie, J.C., Paulus, E., Suryani, M., Hadi, S., Verleysen, M., Chhun, S., Ogier, J.M. (2018). *ICFHR 2018 competition on document image analysis tasks for Southeast Asian palm leaf manuscripts.

---

📚 Citation

If you find PALM-LAY interesting and useful for your research, please cite:

@inproceedings{thuon2025palmlay,
  title     = {PALM-LAY: A Multi-Script Cross-Regional Dataset for Layout Analysis of Palm Leaf Manuscripts},
  author    = {Thuon, Nimol and Du, Jun and Theang, Panhapin and Thuon, Ratana},
  booktitle = {Proceedings of the International Conference on Document Analysis and Recognition (ICDAR)},
  year      = {2025},
  publisher = {Springer},
}