SQ-LLaVA: A New Visual Instruction Tuning Method that Enhances General-Purpose Vision-Language Understanding and Image-Oriented Question Answering through Visual Self-Questioning