asteroid-team / asteroid

The PyTorch-based audio source separation toolkit for researchers

Home Page:https://asteroid-team.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

depth-wise separable convolution code question

lsg1213 opened this issue · comments

self.shared_block = nn.Sequential(
in_conv1d,
nn.PReLU(),
conv_norm(hid_chan),
depth_conv1d,
nn.PReLU(),
conv_norm(hid_chan),
)

In this code, the order of operation is point-wise -> depth-wise convolution
However, in my knowledge, the order of the operation is depth-wise -> point-wise convolution.
May I ask you the reason that the code was made like that?

Oh, the paper writer made the model like that
I didn't check before to ask

  • I read the code wrong
    I'm sorry