Efficient Knowledge Distillation for Lightweight Facial Expression Recognition CNNs
Dedi Wijaya, Dexter Sing Fong Leong, Ho Yin Chan, Yit Hong Choo, Tao Zhou, Andi Prademon Yunus, Xinjie DengNon-Scopus13 Juni 2025Afiliasi: AU, CN, ID
Abstrak
Facial Expression Recognition (FER) plays an important role in human-computer interaction and affective computing. This paper presents a CNN-based student model that uses knowledge distillation to transfer knowledge from a high-capacity teacher model. The teacher model is based on EfficientNet augmented with attention mechanisms, while the CNN-based student model has a more concise architecture. The proposed approach is evaluated on the Real-world Affective Faces Database (RAF-DB) dataset, which contains seven emotion classes. Experimental results demonstrate that the distilled student model achieves competitive accuracy while significantly reducing computational complexity, making it suitable for real-time applications.
Identifier
- DOI: https://doi.org/10.1109/prml66062.2025.11159812
- OpenAlex: https://openalex.org/W4414271698

