LightweightGAN/DepthWiseConv2d.py

from torch import nn


class DepthWiseConv2d(nn.Module):
    def __init__(self, dim_in, dim_out, kernel_size, padding=0, stride=1, bias=True):
        super().__init__()
        self.net = nn.Sequential(
            nn.Conv2d(
                dim_in,
                dim_in,
                kernel_size=kernel_size,
                padding=padding,
                groups=dim_in,
                stride=stride,
                bias=bias,
            ),
            nn.Conv2d(dim_in, dim_out, kernel_size=1, bias=bias),
        )

    def forward(self, x):
        return self.net(x)