What "Not" to Detect: Negation-Aware VLMs via Structured Reasoning and Token Merging

Open in new window