Face-LLaVA is a novel multimodal large language model that analyzes facial expressions and attributes through instruction tuning while generating natural language descriptions for reasoning.

• Creates comprehensive facial analysis through in-context learning
• Developed using FaceInstruct-1M, a specialized face-centered dataset
• Generates natural language descriptions that enable facial reasoning
• Delivers deeper understanding of facial communication for AI systems

Medical Impact: Face-LLaVA offers transformative potential for psychiatry, psychology, and patient monitoring by enabling objective facial expression recognition for pain assessment and mental health diagnosis, creating new opportunities for more effective, personalized care.

Face-LLaVA: Facial Expression and Attribute Understanding through Instruction Tuning