Search papers, labs, and topics across Lattice.
The paper introduces RebusBench, a new benchmark designed to evaluate the cognitive visual reasoning abilities of Large Vision-Language Models (LVLMs) when solving rebus puzzles. Rebus puzzles require LVLMs to extract visual and textual attributes, retrieve linguistic prior knowledge, and perform abstract mapping to synthesize a meaning beyond the pixel space. Experiments on state-of-the-art LVLMs reveal a significant deficiency, with performance saturating below 10% Exact Match and 20% semantic accuracy, indicating a lack of cognitive reasoning despite possessing the necessary visual and linguistic components.
Despite advances in vision-language models, they still fail at rebus puzzles, highlighting a critical gap in cognitive visual reasoning that neither scaling nor in-context learning can fix.
Large Vision-Language Models (LVLMs) have achieved remarkable proficiency in explicit visual recognition, effectively describing what is directly visible in an image. However, a critical cognitive gap emerges when the visual input serves only as a clue rather than the answer. We identify that current models struggle with the complex, multi-step reasoning required to solve problems where information is not explicitly depicted. Successfully solving a rebus puzzle requires a distinct cognitive workflow: the model must extract visual and textual attributes, retrieve linguistic prior knowledge (such as idioms), and perform abstract mapping to synthesize these elements into a meaning that exists outside the pixel space. To evaluate this neurosymbolic capability, we introduce RebusBench, a benchmark of 1,164 puzzles designed to test this specific integration of perception and knowledge. Our evaluation of state-of-the-art models (including Qwen, InternVL, and LLaVA) shows a severe deficiency: performance saturates below 10% Exact Match and 20% semantic accuracy, with no significant improvement observed from model scaling or In-Context Learning (ICL). These findings suggest that while models possess the necessary visual and linguistic components, they lack the cognitive reasoning glue to connect them. Project page available at https://amirkasaei.com/rebusbench/.
