3

I was looking through the architecture of EfficientnetB0 and noticed DepthwiseConv2D operation. Did some digging and found that there's also a SeparableConv2D. What exactly are these operations?

Madara
  • 328
  • 1
  • 5
  • 19
  • 1
    Separable Convolutions consists of Depthwise and Pointwise Convolutions. They are used to reduce the number of trainable parameters ensuring speed and efficiency. They form the core of MobileNets. See [this](https://towardsdatascience.com/a-basic-introduction-to-separable-convolutions-b99ec3102728) blog. – Shubham Panchal May 23 '20 at 05:50

1 Answers1

0

DepthwiseConv2d performs the first step in a depthwise spatial convolution, apply convolution operation on each input channel separately.

reza_gp
  • 1
  • 1
    This does not look like a complete Answer to the OP's question. Can you edit to expand a bit more on it? The second half of your sentence seems incomplete, has a grammatical errors, or is difficult to understand. Please explain the difference between both. It would also be helpful to include links to documentation so visitors could follow up for more info. – SherylHohman Apr 05 '21 at 02:44