MixConv: Mixed Depthwise Convolutional Kernels

22 Jul 2019Mingxing TanQuoc V. Le

Depthwise convolution is becoming increasingly popular in modern efficient ConvNets, but its kernel size is often overlooked. In this paper, we systematically study the impact of different kernel sizes, and observe that combining the benefits of multiple kernel sizes can lead to better accuracy and efficiency... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification ImageNet MixNet-S Top 1 Accuracy 75.8% # 114
Top 5 Accuracy 92.8% # 79
Number of params 4.1M # 68
Image Classification ImageNet MixNet-M Top 1 Accuracy 77% # 103
Top 5 Accuracy 93.3% # 72
Number of params 5M # 65
Image Classification ImageNet MixNet-L Top 1 Accuracy 78.9% # 78
Top 5 Accuracy 94.2% # 60
Number of params 7.3M # 53

Methods used in the Paper