SemVink: Advancing VLMs' Semantic Understanding of Optical Illusions via Visual Global Thinking Paper • 2506.02803 • Published Jun 3 • 1