I'm thrilled to share that our paper, KRETA: Korean Reading and rEasoning in Text-rich VQA Attuned to diverse visual contexts, has been accepted to the EMNLP 2025 Main Conference.
This year’s acceptance rate was highly competitive—only 22.16% of 8,174 submissions were selected as Main papers. KRETA introduces a comprehensive benchmark for evaluating Korean reading and reasoning in text-rich Visual Question Answering, spanning 15 domains and 26 image types.
Furthermore, we built a semi-automatic VQA generation pipeline optimized for text-rich environments to ensure data quality. We hope KRETA will serve as a strong foundation for advancing multilingual vision-language models and Korean NLP research.
As a co-author, I'm excited to contribute to advancing multilingual vision-language model research.
