SemVink: Advancing VLMs' Semantic Understanding of Optical Illusions via Visual Global Thinking
This reveals a critical architectural flaw in VLMs that limits their real-world robustness for applications like medical imaging and security, though the solution is incremental.
The paper tackles the problem that vision-language models (VLMs) fail to detect hidden content in optical illusions or AI-generated images, achieving near-zero accuracy (0-5.36%) on a new benchmark (HC-Bench), and shows that simply scaling images to low resolutions (32-128 pixels) unlocks >99% accuracy by eliminating visual noise.
Vision-language models (VLMs) excel in semantic tasks but falter at a core human capability: detecting hidden content in optical illusions or AI-generated images through perceptual adjustments like zooming. We introduce HC-Bench, a benchmark of 112 images with hidden text, objects, and illusions, revealing that leading VLMs achieve near-zero accuracy (0-5.36%)-even with explicit prompting. Humans resolve such ambiguities instinctively, yet VLMs fail due to an overreliance on high-level semantics. Strikingly, we propose SemVink (Semantic Visual Thinking) by simply scaling images to low resolutions (32-128 pixels), which unlocks >99% accuracy by eliminating redundant visual noise. This exposes a critical architectural flaw: VLMs prioritize abstract reasoning over low-level visual operations crucial for real-world robustness. Our work urges a shift toward hybrid models integrating multi-scale processing, bridging the gap between computational vision and human cognition for applications in medical imaging, security, and beyond.