SynthTIGER: Synthetic Scene Text Image Generation for OCR Recognition

SynthTIGER is a synthetic text image generator developed by Clova AI for the ICDAR 2021 conference. It enables the creation of large-scale scene text datasets for training and evaluating deep learning-based OCR recognition models.

Use Cases

• Training deep learning models for scene-text-recognition using generated text labels
• Benchmarking OCR-recognition performance across different font styles and transformations
• Data augmentation for scene-text tasks by generating variations of rare characters or scripts

Strengths

• ICDAR 2021 peer-reviewed methodology
• MIT licensed for open research and commercial use
• Highly customizable generation parameters for fonts and backgrounds

Limitations

• Synthetic data may not fully capture real-world sensor noise or environmental artifacts
• Requires local execution of the generation pipeline to produce the actual dataset

Provenance

Source

Clova AI (ICDAR 2021)

Collection Method

synthetic

Freshness

Last updated June 2024; based on 2021 research.