pepder/thai-ocr-text-images

Thai OCR Text Images (Test)

Description

Thai OCR Text Images is a dataset of Thai text images designed for Optical Character Recognition (OCR) and Vision-Language Model (VLM) training.

Each sample consists of:

• A cropped image containing Thai text.
• The corresponding ground truth text.

The images are generated from PDF documents and are intended to improve Thai text recognition performance in machine learning models.

---

Dataset Structure

thai-ocr-text-images/
├── README.md
└── data/
    ├── metadata.csv
    └── images/
        ├── pdf_1.png
        ├── pdf_2.png
        └── ...

Example metadata.csv

file_name	text	category
images/pdf_1.png	มิถุนายน	document_th
images/pdf_2.png	๒๕๖๑	document_th

---

Data Fields

Field	Description
file_name	Relative path to the image file
text	Ground truth Thai text
category	Category or source type of the text image

---

Intended Uses

This dataset can be used for:

• Thai OCR
• OCR fine-tuning
• Vision Language Models (VLM)
• Document AI
• Image-to-Text models
• Benchmarking Thai OCR systems

---

Source

The images are extracted and processed from Thai PDF documents.

---

Citation

If you use this dataset in your research or project, please cite:

Thai OCR Text Images Dataset
https://huggingface.co/datasets/pepder/thai-ocr-text-images