QEA-Net: Quantum-Effects-based Attention Networks

In the past decade, the attention mechanism has played an increasingly important role in computer vision. Such an attention mechanism can be regarded as a dynamic weight adjustment process based on features of the input image. In this paper, we propose Quantum-Effects-based Attention Networks (QEA-Net), the simple yet effective attention networks, they can be integrated into many network architectures seamlessly. QEA-Net uses quantum effects between two identical particles to enhance the global channel information representation of the attention module. Our method could consistently outperform the SENet, with a lower number of parameters and computational cost. We evaluate QEA-Net through experiments on ImageNet-1K and compare it with state-of-the-art counterparts. We also demonstrated the effect of QEA-Net in combination with pre-trained networks on small downstream transfer learning tasks.