Khmer and English OCR Training Dataset with 12 Million Synthetic Text Line Images

12,138,214 synthetic image-text pairs form a large-scale resource for training optical character recognition models. The dataset, created by author mrrtmob, contains high-quality images of text lines in Khmer, English, and mixed languages. It was last updated on the Hugging Face platform in January 2026.

Use Cases

• Train OCR models for Khmer script recognition based on the high volume of Khmer text images.
• Develop multilingual OCR systems based on the inclusion of English and mixed-language text lines.
• Benchmark synthetic data generation techniques for OCR based on the dataset's large-scale synthetic nature.
• Fine-tune pre-trained vision-language models on text-in-the-wild tasks based on the image-text pair format.

Strengths

• Contains 12,138,214 high-quality synthetic images, providing substantial scale for model training.
• Supports two distinct languages (Khmer and English) and their mixtures, enabling multilingual applications.
• Provides ground truth text strings paired with each image, which is essential for supervised OCR training.

Limitations

• The dataset is synthetic, which may limit its realism and generalizability to real-world document images.
• Column-level documentation is absent; field semantics must be inferred after download.
• Freshness should be verified as the last update timestamp is in the future (2026-01-19).

Provenance

Source

huggingface

Collection Method

Synthetically generated.

Freshness

Last updated 2026-01-19 15:58:42