Not working with conditional StyleGAN2-ADA model

[sriderya]

Hello,

I trained a conditional StyleGAN2-ada-pytorch model with a custom dataset. Then, I converted my pkl model to pt model by using https://github.com/dvschultz/stylegan2-ada-pytorch/blob/main/export_weights.py. Then, I integrated my model to Ganspace wrapper and I ran it. This gives me the following error.

RuntimeError: Error(s) in loading state_dict for Generator:
	size mismatch for style.1.weight: copying a param with shape torch.Size([512, 1024]) from checkpoint, the shape in current model is torch.Size([512, 512]).
	size mismatch for convs.6.conv.weight: copying a param with shape torch.Size([1, 256, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 512, 512, 3, 3]).
	size mismatch for convs.6.activate.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]).
	size mismatch for convs.7.conv.weight: copying a param with shape torch.Size([1, 256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 512, 512, 3, 3]).
	size mismatch for convs.7.conv.modulation.weight: copying a param with shape torch.Size([256, 512]) from checkpoint, the shape in current model is torch.Size([512, 512]).
	size mismatch for convs.7.conv.modulation.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]).
	size mismatch for convs.7.activate.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]).
	size mismatch for convs.8.conv.weight: copying a param with shape torch.Size([1, 128, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 256, 512, 3, 3]).
	size mismatch for convs.8.conv.modulation.weight: copying a param with shape torch.Size([256, 512]) from checkpoint, the shape in current model is torch.Size([512, 512]).
	size mismatch for convs.8.conv.modulation.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]).
	size mismatch for convs.8.activate.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]).
	size mismatch for convs.9.conv.weight: copying a param with shape torch.Size([1, 128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 256, 256, 3, 3]).
	size mismatch for convs.9.conv.modulation.weight: copying a param with shape torch.Size([128, 512]) from checkpoint, the shape in current model is torch.Size([256, 512]).
	size mismatch for convs.9.conv.modulation.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]).
	size mismatch for convs.9.activate.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]).
	size mismatch for convs.10.conv.weight: copying a param with shape torch.Size([1, 64, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 128, 256, 3, 3]).
	size mismatch for convs.10.conv.modulation.weight: copying a param with shape torch.Size([128, 512]) from checkpoint, the shape in current model is torch.Size([256, 512]).
	size mismatch for convs.10.conv.modulation.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]).
	size mismatch for convs.10.activate.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
	size mismatch for convs.11.conv.weight: copying a param with shape torch.Size([1, 64, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([1, 128, 128, 3, 3]).
	size mismatch for convs.11.conv.modulation.weight: copying a param with shape torch.Size([64, 512]) from checkpoint, the shape in current model is torch.Size([128, 512]).
	size mismatch for convs.11.conv.modulation.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
	size mismatch for convs.11.activate.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
	size mismatch for to_rgbs.3.conv.weight: copying a param with shape torch.Size([1, 3, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 3, 512, 1, 1]).
	size mismatch for to_rgbs.3.conv.modulation.weight: copying a param with shape torch.Size([256, 512]) from checkpoint, the shape in current model is torch.Size([512, 512]).
	size mismatch for to_rgbs.3.conv.modulation.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]).
	size mismatch for to_rgbs.4.conv.weight: copying a param with shape torch.Size([1, 3, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 3, 256, 1, 1]).
	size mismatch for to_rgbs.4.conv.modulation.weight: copying a param with shape torch.Size([128, 512]) from checkpoint, the shape in current model is torch.Size([256, 512]).
	size mismatch for to_rgbs.4.conv.modulation.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]).
	size mismatch for to_rgbs.5.conv.weight: copying a param with shape torch.Size([1, 3, 64, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 3, 128, 1, 1]).
	size mismatch for to_rgbs.5.conv.modulation.weight: copying a param with shape torch.Size([64, 512]) from checkpoint, the shape in current model is torch.Size([128, 512]).
	size mismatch for to_rgbs.5.conv.modulation.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).

I believe that the mismatches arise from conditioning. It looks like something changes after label embeddings are added to unconditional Stylegan2-ada-pytorch network. Is there any solution for this problem or any suggestion to overcome this?

Thanks in advance

[MikhailKulyabin]

I have the same problem

[HashimotoJun25]

I have the same problem

[marceljhuber]

Hi,
I am currently facing the same problem as you with regards to style1.weight:

RuntimeError: Error(s) in loading state_dict for Generator:
size mismatch for style.1.weight: copying a param with shape torch.Size([512, 1024]) from checkpoint, the shape in current model is torch.Size([512, 512]).

After reading every related problem I'm still not sure why this vector contains 1024 and how I should handle this. I don't think it's because of the auto-config because I trained another model using paper256 and the first mapping layer has again shape [512, 1024]. Could anyone explain to me at which point in training I could change this? And additionally, how I can use/convert my model containing [512, 1024]-vectors?
Best wishes,
Marcel!