Cii Bench: Chinese Image Metaphor Understanding for MLLMs

Cii Bench is a vision-and-language benchmark created by MING-ZCH for the ACL 2025 conference to evaluate how Multimodal Large Language Models (MLLMs) interpret metaphorical meanings in Chinese imagery. The dataset provides a specialized framework for testing deep semantic understanding beyond literal object recognition, focusing on culturally specific visual cues.

Use Cases

• Testing MLLM accuracy in identifying 'deep implications' within Chinese visual metaphors
• Benchmarking 'vision-and-language' alignment for complex semantic tasks
• Evaluating model performance on culturally-specific 'metaphor' interpretation

Strengths

• Peer-reviewed benchmark accepted for ACL 2025
• Focuses on high-level metaphorical reasoning rather than literal object detection
• Addresses cultural nuances specific to Chinese visual communication

Limitations

• Total record count and class distribution are not publicly specified in the repository metadata
• Geographic and cultural scope is restricted to Chinese contexts
• Lack of explicit license information may restrict commercial usage

Provenance

Source

MING-ZCH (ACL 2025 publication)

Collection Method

Expert-curated benchmark for academic evaluation

Freshness

Last updated in October 2025, coinciding with the ACL 2025 publication cycle.

Geography

China