FR-CBAMNet: A Face Recognition Algorithm by Fusion of Transformer and CNN

With the development of deep learning, convolutional neural network is mainstream in face recognition but has two problems: poor recognition accuracy due to neglecting global semantic and key feature information, and difficulty in deployment on resource-constrained devices. This paper proposes lightweight face recognition algorithm FR-CBAMNet, combining Transformer and CNN. It reduces parameters with deeply separable convolution, introduces CBAM for feature selection, and integrates Transformer to capture global semantic information. Experimentally evaluated on nine validation sets, FR-CBAMN et with only 6.1 × 10⁵ parameters achieves good balance between parameters and performance.