How to implement 2D depthwise/seperable convolution layer in 2017b?

Lucas Katzmann

2019 10 28

0 个回答

10 次查看（30 天）

0 个投票

Hi!

I want to know, if anyone has an idea how to create a custom depthwise or seperable convolutionlayer with Matlab with the methods described here.

We have PCT and MDCS Licenses for 2017b and therefore I cannot use 2019b, where those layers are available. I was now thinking about implementing a custom layer which perfoms depthwise and pointwise convolution like the corespondig Keras layer seperable_conv2d. But at the moment I have no idea how to start.

I have sensordata timeseries and I want one kernel convolving over dimension "time" and the other convolving over dimension "sensors".

Can anybody help me?

Thanks!