Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications
This work addresses the problem of improving accuracy in compact neural networks for mobile applications, representing an incremental advancement over existing methods.
The paper tackles the performance degradation in compact neural networks due to blocked inter-group information exchange in sparsely-connected convolutions by introducing merging and evolution operations and the ME module, resulting in MENet models that outperform state-of-the-art compact networks by up to 4.1% on benchmarks like ILSVRC 2012 and PASCAL VOC 2007 under specific computational budgets.
Compact neural networks are inclined to exploit "sparsely-connected" convolutions such as depthwise convolution and group convolution for employment in mobile applications. Compared with standard "fully-connected" convolutions, these convolutions are more computationally economical. However, "sparsely-connected" convolutions block the inter-group information exchange, which induces severe performance degradation. To address this issue, we present two novel operations named merging and evolution to leverage the inter-group information. Our key idea is encoding the inter-group information with a narrow feature map, then combining the generated features with the original network for better representation. Taking advantage of the proposed operations, we then introduce the Merging-and-Evolution (ME) module, an architectural unit specifically designed for compact networks. Finally, we propose a family of compact neural networks called MENet based on ME modules. Extensive experiments on ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet consistently outperforms other state-of-the-art compact networks under different computational budgets. For instance, under the computational budget of 140 MFLOPs, MENet surpasses ShuffleNet by 1% and MobileNet by 1.95% on ILSVRC 2012 top-1 accuracy, while by 2.3% and 4.1% on PASCAL VOC 2007 mAP, respectively.