Uncertain-aware patch learning for ambiguous facial expression recognition

Despite substantial progress in Facial Expression Recognition (FER) in the past few years, many methods have been developed to recognize unreliable facial expressions. Occlusions lead to incomplete facial regions and contaminated facial features in real-world scenarios, which undoubtedly hinders the performance of deep learning models for FER in the wild. Therefore, a new uncertain-aware patch learning method (UAPL) is proposed to improve the robustness of deep models against the ambiguous FER problem. Firstly, we leverage the topological information of the labels from related but different tasks. Because facial information should have similar expression distributions to their neighbours in the label space. Then, the backbone convolutional neural network is used to extract facial feature maps from expression and their related facial features, which are cropped into multiple regional patches to extract. Next, the patch attention is designed to perceive occluded regions by adaptively calculating the patch-level attention weights of local features, improving confidence in assisted labels. Finally, optimize the entire facial recognition model through task guidance loss. Experiments are conducted on two widely used expression datasets, our proposed method is evaluated on various datasets and consistently outperforms state-of-the-art methods with a huge margin.