How lambda layer handle the downsample in LambdaResNet?

[qiaoran-dawnlight]

Hi,
Thanks for your clear code, i try to implement the LambdaResNet.
Does lambda layer replace all conv2d layer?
If so, how does lambda layer handle the downsample in conv2d, like stride=2?
Or just keep the conv2d if stride =2, replace only the conv2d layers in stride =1?

[PistonY]

Paper already tell you
We apply 3x3 average-pooling with stride 2 after the lambda layers to downsample in place of the strided convolution.

[qiaoran-dawnlight]

@PistonY Noted ,Thanks