Benchmark
MMStar
A benchmark specifically designed to assess the degree of visual dependency in multimodal large language models. It curates questions that cannot be answered correctly through language priors alone, thereby evaluating a model's genuine visual understanding capability rather than reliance on statistical language patterns.
Timeline (last 90 days)
Oct 9
Benchmarks such as MMVP and MMStar are specifically designed to detect the degree of visual dependency, deliberately including questions that cannot be answered correctly through language priors alone.
Unverified50%