1x1convolution

[Vegetabhl]

This line of code in splatconv2d self.conv = Conv2d(in_channels, channels radix, kernel_size, stride, padding, dilation, groups=group radix, bias=bias, **kwargs)

My understanding is that the input feature map is divided into {group * radius}，Each {radix} is a group，According to the principle of equivalence, there will be 3x3 convolution in each group(kernel_size=3)，But I can't find the 1x1 convolution in the group on the graph in the paper? I don't seem to find it in the code?

[zhanghang1989]

There is a conv 1x1 after concatenation:

[Vegetabhl]

First of all, thank you for your reply。In splatconv2d，I looked at the code, which seems to correspond to this figure。But I didn't find the 1 * 1 convolution marked on the figure in the code。

[zhanghang1989]

That figure shows the entire bottleneck block: https://github.com/zhanghang1989/ResNeSt/blob/master/resnest/torch/models/resnet.py#L41