GhostNet: More Features from Cheap Operations

Deploying convolutional neural networks (CNNs) on embedded devices is difficult due to the limited memory and computation resources. The redundancy in feature maps is an important characteristic of those successful CNNs, but has rarely been investigated in neural architecture design. This paper proposes a novel Ghost module to generate more feature maps from cheap operations. Based on a set of intrinsic feature maps, we apply a series of linear transformations with cheap cost to generate many ghost feature maps that could fully reveal information underlying intrinsic features. The proposed Ghost module can be taken as a plug-and-play component to upgrade existing convolutional neural networks. Ghost bottlenecks are designed to stack Ghost modules, and then the lightweight GhostNet can be easily established. Experiments conducted on benchmarks demonstrate that the proposed Ghost module is an impressive alternative of convolution layers in baseline models, and our GhostNet can achieve higher recognition performance (e.g. $75.7\%$ top-1 accuracy) than MobileNetV3 with similar computational cost on the ImageNet ILSVRC-2012 classification dataset. Code is available at https://github.com/huawei-noah/ghostnet

PDF Abstract CVPR 2020 PDF CVPR 2020 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Classification ImageNet GhostNet ×1.0 Top 1 Accuracy 73.9% # 907
Number of params 5.2M # 410
GFLOPs 0.141 # 9
Image Classification ImageNet GhostNet ×1.3 Top 1 Accuracy 75.7% # 864
Number of params 7.3M # 456
GFLOPs 0.226 # 18
Image Classification ImageNet GhostNet ×0.5 Top 1 Accuracy 66.2% # 964
Number of params 2.6M # 362
GFLOPs 0.042 # 1
Image Classification ImageNet Ghost-ResNet-50 (s=4) Top 1 Accuracy 74.1% # 904
Number of params 6.5M # 444
GFLOPs 1.2 # 114
Image Classification ImageNet Ghost-ResNet-50 (s=2) Top 1 Accuracy 75% # 885
Number of params 13M # 505
GFLOPs 2.2 # 154

Methods