arXiv AI
Beyond What Meets the Eye: Unveiling Situational Illusions for Multimodal Large Language Models
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Aug 26
Seeing vs. Believing: Evaluating the Language Bias of Open-Source MLLMs in Counter-Intuitive Scenes
The paper introduces CAIT, a benchmark of 400 synthetic scenes featuring counter‑intuitive actions that challenge multimodal large language models (MLLMs). Human participants and proprietary models like Claude and Gemini perform well, but standard open‑source instruction‑tuned MLLMs fail, largely due to a strong language prior that overrides contradictory visual evidence. The study shows that Chain‑of‑Thought reasoning can help but introduces new issues, while targeted fine‑tuning and structured prompting can reduce reliance on language priors and improve visual grounding.
arXiv AI
Aug 7
HoloCount: A Holistic Visual Counting Benchmark for MLLMs
arXiv:2607. 06420v1 Announce Type: cross Abstract: Visual counting is a fundamental pillar of multimodal intelligence, requiring a seamless integration of fine-grained grounding and spatial reasoning.
arXiv AI
Aug 28
Can LVLMs Uncover the Truth Behind Visual Illusions? An Analysis of Perceptual and Reasoning Capabilities
The paper introduces Illusion-Reasoning, a benchmark that uses real-world visual illusion images paired with annotated question-answer pairs to evaluate Large Vision Language Models (LVLMs). It argues that current evaluations focus too narrowly on perception or specific domains, and that a joint assessment of perception and reasoning in open-world settings is lacking. Using this benchmark, the authors demonstrate that many LVLMs’ reasoning abilities are less advanced than previously claimed, offering new insights and directions for optimization.
arXiv AI
Jul 17
Contextualized Evaluation of Vision Language Models through Dynamic, Multi-turn Interactions
arXiv:2607. 14499v1 Announce Type: new Abstract: Multi-modal Large Language Models (MLLMs) have made substantial advances on benchmarks, yet their real-world effectiveness remains uncertain.
arXiv Computer Vision
Sep 1
Think, Look, and Revise: Inconsistency-Aware Visual Self-Correction in MLLMs
arXiv:2608.29374v1 Announce Type: new Abstract: Tool-augmented multimodal reasoning integrates external tools (e.g., object detection, depth estimation) into multimodal large language models (MLLMs)...
arXiv AI
Sep 7
When Seeing Overrides Knowing: Visual Dominance and Deferral-Based Method for Personalized Safety in VLMs
The paper introduces MPS-Bench, a benchmark of 5,181 scenarios from 584 real-world images across 12 high-risk domains, each paired with a hidden user profile, to evaluate personalized safety in vision‑language models (VLMs). Eight leading VLMs were tested and found to almost always respond directly (86‑99%) without seeking missing context, scoring no higher than 2.6/5 on personalized safety. The authors identify a phenomenon called visual dominance, where visual information enters text representations early and suppresses textual risk signals, and propose PRISM, a lightweight input monitor that predicts when a query should be deferred, achieving 0.978 AUC and outperforming all tested models on the safety‑utility Pareto frontier.