I have a question about "Get_Weight_Gradient" function in content aware pruning.
You only consider the gradient of to_rgbs[-1] in rgb layers.
(I mean this code "module_list = [g_module.conv1] + list(g_module.convs) + [g_module.to_rgbs[-1]]")
However, there are many to_rgbs layers in stylegan2
(e.g., to_rgbs[0], to_rgbs[1], to_rgbs[2], ...)
thank you for sharing the your code.
I have a question about "Get_Weight_Gradient" function in content aware pruning.
You only consider the gradient of to_rgbs[-1] in rgb layers. (I mean this code "module_list = [g_module.conv1] + list(g_module.convs) + [g_module.to_rgbs[-1]]")
However, there are many to_rgbs layers in stylegan2 (e.g., to_rgbs[0], to_rgbs[1], to_rgbs[2], ...)
I think that we consider all to_rgbs layer.
Why you compute the only to_rgbs[-1]?