VirtualLUO/

MMTIT_Bench

A Multilingual and Multi-Scenario Benchmark with Cognition–Perception–Reasoning Guided Text-Image Machine Translation

中文版 • Paper • GitHub • HuggingFace

MMTIT-Bench is a human-verified benchmark for end-to-end Text-Image Machine Translation (TIMT). It contains 1,400 images spanning 14 non-English and non-Chinese languages across diverse real-world scenarios, with bilingual (Chinese & English) translation annotations. We also propose… See the full description on the dataset page: https://huggingface.co/datasets/VirtualLUO/MMTIT_Bench.