VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models

Open in new window