Fetching the paper…

NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples · Around