NaturalBench is a robust, vision-centric, bias-free VQA benchmark for evaluating compositional reasoning in visual language models(VLMs). A whole thread about NaturalBench and why it is an important benchmark in multimodal community

→ View original post on X — @jeande_d,