| Abstract [eng] |
Facial expression recognition in the wild remains a challenging computer vision task because facial images captured in unconstrained conditions are affected by pose variation, illumination changes, blur, occlusion, background clutter, and substantial intra-class variability. In addition, large-scale inthe-wild datasets such as AffectNet-8 exhibit strong class imbalance, which makes it insufficient to evaluate performance using overall accuracy alone. These challenges motivate the development of methods that are not only accurate, but also efficient and stable enough for practical real-time use. |
