Abstract
Facial Expression Recognition (FER) is a machine learning problem that dealswith recognizing human facial expressions. While existing work has achievedperformance improvements in recent years, FER in the wild and under challengingconditions remains a challenge. In this paper, a lightweight patch andattention network based on MobileNetV1, referred to as PAtt-Lite, is proposedto improve FER performance under challenging conditions. A truncatedImageNet-pre-trained MobileNetV1 is utilized as the backbone feature extractorof the proposed method. In place of the truncated layers is a patch extractionblock that is proposed for extracting significant local facial features toenhance the representation from MobileNetV1, especially under challengingconditions. An attention classifier is also proposed to improve the learning ofthese patched feature maps from the extremely lightweight feature extractor.The experimental results on public benchmark databases proved the effectivenessof the proposed method. PAtt-Lite achieved state-of-the-art results on CK+,RAF-DB, FER2013, FERPlus, and the challenging conditions subsets for RAF-DB andFERPlus. The source code for the proposed method will be available athttps://github.com/JLREx/PAtt-Lite.