Is it a True Conv Free Method ?

[kabbas570]

Hello, Thanks for the excellent work,
I was just wondering about the claims made by the paper 'Convolution Free', although I can see two convolutional layers in the model architecture. Is there any explanation for this?

(

transdeeplab/model/backbones/swin.py

Line 34 in c2069c0

self.proj = nn.Conv2d(in_chans, embed_dim,

)

transdeeplab/model/decoder.py

Line 173 in c2069c0

self.output = nn.Conv2d(int(input_dim)*2, num_classes, kernel_size=1, bias=False)

Thanks
Cheers
Abbas