SemVink
Visual global thinking for VLM semantic understanding of optical illusions.
SemVink investigates why vision-language models struggle with hidden semantic content in optical illusions and how visual global thinking can bridge this gap.
- Project page: johnnyzeppelin.github.io/vlm-semvink
- ACL Anthology: 2025.emnlp-main.1381
- Venue: EMNLP 2025 Main Conference, oral presentation