RepGhost: A Hardware-Efficient Ghost Module via Re-parameterization
This work addresses the problem of hardware inefficiency in lightweight CNN design for mobile applications, offering an incremental improvement over existing methods like GhostNet and MobileNetV3.
The paper tackles the computational inefficiency of feature reuse via concatenation in lightweight CNNs by proposing RepGhost, a hardware-efficient module using reparameterization for implicit feature reuse, resulting in RepGhostNet achieving 2.5% higher Top-1 accuracy than GhostNet 0.5x on ImageNet with fewer parameters and comparable latency on ARM mobile devices.
Feature reuse has been a key technique in light-weight convolutional neural networks (CNNs) architecture design. Current methods usually utilize a concatenation operator to keep large channel numbers cheaply (thus large network capacity) by reusing feature maps from other layers. Although concatenation is parameters- and FLOPs-free, its computational cost on hardware devices is non-negligible. To address this, this paper provides a new perspective to realize feature reuse implicitly and more efficiently instead of concatenation. A novel hardware-efficient RepGhost module is proposed for implicit feature reuse via reparameterization, instead of using concatenation operator. Based on the RepGhost module, we develop our efficient RepGhost bottleneck and RepGhostNet. Experiments on ImageNet and COCO benchmarks demonstrate that our RepGhostNet is much more effective and efficient than GhostNet and MobileNetV3 on mobile devices. Specially, our RepGhostNet surpasses GhostNet 0.5x by 2.5% Top-1 accuracy on ImageNet dataset with less parameters and comparable latency on an ARM-based mobile device. Code and model weights are available at https://github.com/ChengpengChen/RepGhost.