to_rgb branch has only 1 learnable kernel about gigagan-pytorch HOT 7 CLOSED

this doesnt fix the issue i'm talking about:
to_rgb = AdaptiveConv2DMod(dim_out, channels, 1, num_conv_kernels = 1, demod = False)
num_conv_kernels is always 1 for rgb regardless of Generator setting

from gigagan-pytorch.

but you still compute kernel weights for rgb branch

style_embed_split_dims.extend([
dim_in, # for first conv in resnet block
dim_kernel_mod, # first conv kernel selection
dim_out, # second conv in resnet block
dim_kernel_mod, # second conv kernel selection
dim_out, # to RGB conv
dim_kernel_mod, # RGB conv kernel selection
])

from gigagan-pytorch.

but i agree that this path looks super convoluted :) i would rather move mod+kernel_mod projection inside AdaptiveConv2DMod

from gigagan-pytorch.

i may move all that modulations into the adaptive conv2d mod at some point

its too confusing the way i have it currently

from gigagan-pytorch.

@inspirit yup, that is intentional, based on point 2 in #33

from gigagan-pytorch.

@inspirit 🤦‍♂️ yes you are right

from gigagan-pytorch.

@inspirit yes i'll get around to it 😅

from gigagan-pytorch.