Zero-Shot Emotion Detection with LLMs
Using GPT-4o-mini to automatically annotate facial emotions in real-world scenarios
This research demonstrates how large language models can be leveraged for zero-shot facial emotion annotation in everyday video contexts, requiring no prior training on facial emotion datasets.
- Employs GPT-4o-mini to analyze key video frames and classify emotions into seven categories
- Evaluates performance on the DailyLife subset of the FERV39k dataset
- Offers a scalable approach to emotion recognition without specialized emotion recognition systems
Security Applications: This technology has significant implications for threat detection, surveillance systems, and behavioral analysis in security contexts, enabling automated emotional state assessment in security screening and monitoring.