QLIP: A Dynamic Quadtree Vision Prior Enhances MLLM Performance Without Retraining

Open in new window