Adaptive Multilayer Perceptual Attention Network for Facial Expression Recognition

In complex real-world situations, problems such as illumination changes, facial occlusion, and variant poses make expression recognition (FER) a challenging task. To solve the robustness problem, this paper proposes an adaptive multilayer perceptual attention network (AMP-Net) that is inspired by attributes perception mechanism of human visual system. AMP-Net extracts global, local, salient emotional features with different fine-grained to learn underlying diversity key information emotions. Different from existing methods, can adaptively guide focus on multiple finer distinguishable local patches occlusion poses, improving effectiveness learning potential information. addition, proposed global module receptive field in domain, also supplements region high emotion correlation based prior knowledge capture texture details avoid important loss. Many experiments show achieves good generalizability state-of-the-art results several datasets, including RAF-DB, AffectNet-7, AffectNet-8, SFEW 2.0, FER-2013, FED-RO, accuracies 89.25%, 64.54%, 61.74%, 61.17%, 74.48%, 71.75%, respectively. All codes training logs are publicly available at https://github.com/liuhw01/AMP-Net .