A Simple Recipe to Boost the Performance of MLLMs on Your Custom Use Case