Attentional Feature Fusion

TL;DR

This paper introduces a unified attentional feature fusion scheme with multi-scale and iterative attention modules, improving feature integration in neural networks and outperforming state-of-the-art models on CIFAR-100 and ImageNet.

Contribution

It proposes a general attentional feature fusion framework with multi-scale and iterative attention modules, enhancing feature integration across scales and semantics.

Findings

Outperforms state-of-the-art on CIFAR-100 and ImageNet

Fewer layers or parameters achieve better results

Attention mechanisms improve feature fusion effectiveness

Abstract

Feature fusion, the combination of features from different layers or branches, is an omnipresent part of modern network architectures. It is often implemented via simple operations, such as summation or concatenation, but this might not be the best choice. In this work, we propose a uniform and general scheme, namely attentional feature fusion, which is applicable for most common scenarios, including feature fusion induced by short and long skip connections as well as within Inception layers. To better fuse features of inconsistent semantics and scales, we propose a multi-scale channel attention module, which addresses issues that arise when fusing features given at different scales. We also demonstrate that the initial integration of feature maps can become a bottleneck and that this issue can be alleviated by adding another level of attention, which we refer to as iterative attentional feature fusion. With fewer layers or parameters, our models outperform state-of-the-art networks on both CIFAR-100 and ImageNet datasets, which suggests that more sophisticated attention mechanisms for feature fusion hold great potential to consistently yield better results compared to their direct counterparts. Our codes and trained models are available online.