A holistic benchmark and toolkit for evaluating and enhancing the cognition-based emotional understanding capabilities of Multimodal Large Language Models (MLLMs) grounded by Theory-of-Mind (ToM).

https://huggingface.co/datasets/Eureka-Leo/Emotion.Intelligence

• /without_prompt: Contains the raw dataset samples for all tasks in JSON format.
• /with_prompt: Contains dataset samples enhanced with ToM-driven prompts for all tasks, also in JSON format.

• /test_code: Contains execution scripts for all models.
• /evaluation_code: Contains evaluation scripts for all tasks.
• /train_code: Contains scripts for generating our Chain-of-Thought (CoT) training data.