Large Vision-Language Models as Emotion Recognizers in Context Awareness