This research introduces a novel two-stage fine-tuning approach using Large Vision-Language Models (LVLMs) to recognize complex human emotions from facial expressions.

• Addresses limitations in current emotion recognition systems by capturing subtle emotional cues
• Employs pre-trained LVLMs first fine-tuned on basic expressions, then on compound expressions
• Demonstrates enhanced accuracy in recognizing nuanced, complex emotional states
• Provides a foundation for more natural human-computer interaction

Security Impact: This technology enables more sophisticated security systems capable of detecting emotional states and behavioral anomalies, enhancing threat detection in surveillance applications and improving public safety monitoring.

Original Paper: Compound Expression Recognition via Large Vision-Language Models