Enhancing the SAM model for efficient regional captioning by introducing a lightweight query-based feature mixer.
