Shift-and-Balance Attention
This is an incremental improvement for deep learning practitioners, addressing sensitivity in attention mechanisms to enhance model accuracy.
The paper tackles the problem of balancing contributions between trunk and attention branches in neural networks by proposing Shift-and-Balance attention, which uses a learned control factor to regulate the attention branch and adds it to the trunk, resulting in significant accuracy improvements over Squeeze-and-Excite and competitive performance with state-of-the-art Dynamic Convolution.
Attention is an effective mechanism to improve the deep model capability. Squeeze-and-Excite (SE) introduces a light-weight attention branch to enhance the network's representational power. The attention branch is gated using the Sigmoid function and multiplied by the feature map's trunk branch. It is too sensitive to coordinate and balance the trunk and attention branches' contributions. To control the attention branch's influence, we propose a new attention method, called Shift-and-Balance (SB). Different from Squeeze-and-Excite, the attention branch is regulated by the learned control factor to control the balance, then added into the feature map's trunk branch. Experiments show that Shift-and-Balance attention significantly improves the accuracy compared to Squeeze-and-Excite when applied in more layers, increasing more size and capacity of a network. Moreover, Shift-and-Balance attention achieves better or close accuracy compared to the state-of-art Dynamic Convolution.