A latency-efficient integration of channel attention for ConvNets.

Park, Woongkyu; Choi, Yeongyu; Mekala, Mahammad Shareef; SangChoi, Gyu; Yoo, Kook-Yeol; Jung, Ho-youl

Abstract

Designing fast and accurate neural networks is becoming essential in various vision tasks. Recently, the use of attention mechanisms has increased, aimed at enhancing the vision task performance by selectively focusing on relevant parts of the input. In this paper, we concentrate on squeeze-and-excitation (SE)-based channel attention, considering the trade-off between latency and accuracy. We propose a variation of the SE module, called squeeze-and-excitation with layer normalization (SELN), in which layer normalization (LN) replaces the sigmoid activation function. This approach reduces the vanishing gradient problem while enhancing feature diversity and discriminability of channel attention. In addition, we propose a latency-efficient model named SELNeXt, where the LN typically used in the ConvNext block is replaced by SELN to minimize additional latency-impacting operations. Through classification simulations on ImageNet-1k, we show that the top-1 accuracy of the proposed SELNeXt outperforms other ConvNeXtbased models in terms of latency efficiency. SELNeXt also achieves better object detection and instance segmentation performance on COCO than Swin Transformer and ConvNeXt for small-sized models. Our results indicate that LN could be a considerable candidate for replacing the activation function in attention mechanisms. In addition, SELNeXt achieves a better accuracy-latency trade-off, making it favorable for real-time applications and edge computing.

Citation

PARK, W., CHOI, Y., MEKALA, M.S., CHOI, G.S., YOO, K.-Y. and JUNG, H.-Y. 2025. A latency-efficient integration of channel attention for ConvNets. Computers, maerials and continua [online], 82(3), pages 3965-3981. Available from: https://doi.org/10.32604/cmc.2025.059966

• Journal Article Type: Article
• Acceptance Date: Jan 15, 2025
• Online Publication Date: Mar 6, 2025
• Publication Date: Mar 31, 2025
• Deposit Date: Mar 27, 2025
• Publicly Available Date: Mar 27, 2025
• Journal: Computers, materials and continua
• Print ISSN: 1546-2218
• Electronic ISSN: 1546-2226
• Publisher: Tech Science Press
• Peer Reviewed: Peer Reviewed
• Volume: 82
• Issue: 3
• Pages: 3965-3981
• DOI: https://doi.org/10.32604/cmc.2025.059966
• Keywords: Attention mechanism; Convolutional neural networks; Image classification; Object detection; Semantic segmentation
• Public URL: https://rgu-repository.worktribe.com/output/2761834