Fine-Tuning Swin Transformer and Multiple Weights Optimality-Seeking for Facial Expression Recognition

Facial expression recognition plays a key role in human-computer emotional interaction. However, human faces real environments are affected by various unfavorable factors, which will result the reduction of accuracy. In this paper, we proposed novel method combines Fine-tuning Swin Transformer and Multiple Weights Optimality-seeking (FST-MWOS) to enhanced performance. FST-MWOS mainly consists two crucial components: (FST) (MWOS). FST takes Large as backbone network obtain multiple groups fine-tuned model weights for homologous data domains hyperparameters configurations, augmentation methods, etc. MWOS greedy strategy was used mine locally optimal generalizations epoch interval each group weights. Then, optimality-seeking utilized global solution. Experiments results on RAF-DB, FERPlus AffectNet datasets show that outperforms state-of-the-art methods.