LCAM: Low-Complexity Attention Module for Lightweight Face Recognition Networks

Abstract

Inspired by the human visual system to concentrate on the important region of a scene, attention modules recalibrate the weights of either the channel features alone or along with spatial features to prioritize informative regions while suppressing unimportant information. However, the floating-point operations (FLOPs) and parameter counts are considerably high when one is incorporating these modules, especially for those with both channel and spatial attentions in a baseline model. Despite the success of attention modules in general ImageNet classification tasks, emphasis should be given to incorporating these modules in face recognition tasks. Hence, a novel attention mechanism with three parallel branches known as the Low-Complexity Attention Module (LCAM) is proposed. Note that there is only one convolution operation for each branch. Therefore, the LCAM is lightweight, yet it is still able to achieve a better performance. Experiments from face verification tasks indicate that LCAM achieves similar or even better results compared with those of previous modules that incorporate both channel and spatial attentions. Moreover, compared to the baseline model with no attention modules, LCAM achieves performance values of 0.84% on ConvFaceNeXt, 1.15% on MobileFaceNet, and 0.86% on ProxylessFaceNAS with respect to the average accuracy of seven image-based face recognition datasets.

Hybrid Spatial-Channel Attention Mechanism for Cross-Age Face Recognition

“…Table 2 displays the superior performance of our model. Our proposed method improves accuracy by 1.95% over IEFP [18] (i.e., 95.82% → 97.77%) and 4.07% over LCAM [19] (i.e., 93.70% → 97.77%), demonstrating its effectiveness. Based on experimental results, it can be observed that while the Low-Complexity Attention Module (LCAM) approach has a slight advantage over past modules that employed channel and spatial attention, our proposed hybrid channel-spatial mechanism offers higher flexibility.…”

“…This method helps to remove age information from facial features and obtain a purer representation of the same. Another method, called Low-Complexity Attention Module (LCAM) [19], uses three parallel branches of the attention mechanism with only one convolutional operation in each branch. It is a lightweight method and shows better performance in face recognition tasks.…”

“…MT-MIM [24] 96.10 1.67 MTLFace [23] 96.23 1.54 IEFP [18] 95.82 1.95 PocketNet [52] 96.78 0.99 LCAM [19] 93.70 4.07 Ours 97.77 -…”