upsnet/models/fcn.py has a conv layer definition with parameter in_channels given twice.

[feedforward]

Hi, for the deform-conv def in fcn.py, parameter for both number of input channels and number of output channels is same. (in_channels)
Is this choice deliberate? can you please explain why.

Location - Line 52 https://github.com/uber-research/UPSNet/blob/master/upsnet/models/fcn.py

Thanks

[YuwenXiong]

We only use out_channel for the last conv layer in this subnet, so yes it is.

[feedforward]

I see, Thanks.