VISA: Group-wise Visual Token Selection and Aggregation via Graph Summarization for Efficient MLLMs Inference

Open in new window