Why only 3x3 kernel size in resnet?

mit-han-lab / once-for-all

[ICLR 2020] Once for All: Train One Network and Specialize it for Efficient Deployment

https://ofa.mit.edu/

MIT License

1.89k stars 333 forks source link

Why only 3x3 kernel size in resnet? #55

Closed tfwang08 closed 1 year ago

Hrayo712 commented 3 years ago

I have the same doubt. Any intuition behind this @han-cai ? @

Bixiii commented 3 years ago

In ResNet the 7x7 convolution can be replaced with three 3x3 convolutions. This is described in the paper "Bag of Tricks for Image Classification with Convolutional Neural Networks". I also found an interesting analysis of this here.