No module named 'torch_utils' in run_convert_from_tf.py

[a656418zz]

Hi, thank you very much for sharing. Due to the limitation of the GPU, I'm going to use "run_convert_from_tf.py" to transpose a 512x512 pre-trained model of stylegan2, which I got from https://catalog.ngc.nvidia.com/orgs/nvidia/teams/research/models/stylegan2 to download "stylegan2-ffhq-512x512.pkl", then I would like to try to conversion, but it failed, reporting the following error code.

Traceback (most recent call last):
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 373, in
main()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 350, in main
unpickled = load_tf_models_file(args.input)
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 50, in load_tf_models_file
return Unpickler(fp).load()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 45, in find_class
return super(Unpickler, self).find_class(module, name)
ModuleNotFoundError: No module named 'torch_utils'

My environment is cu101+torch1.7.1+tensorboard2.3.0
Please remind me if there is any information that is not enough, thanks a lot!

[adriansahlman]

Hey!

It seems like the model you are trying to use is from https://github.com/NVlabs/stylegan2-ada-pytorch and not the original tensorflow repo that the conversion tool was designed for.
I am sorry but these models are not supported by the current script as the repo I linked did not exist when I created this code!

[gzhhhere]

Hi, thank you very much for sharing. Due to the limitation of the GPU, I'm going to use "run_convert_from_tf.py" to transpose a 512x512 pre-trained model of stylegan2, which I got from https://catalog.ngc.nvidia.com/orgs/nvidia/teams/research/models/stylegan2 to download "stylegan2-ffhq-512x512.pkl", then I would like to try to conversion, but it failed, reporting the following error code.

Traceback (most recent call last):
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 373, in
main()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 350, in main
unpickled = load_tf_models_file(args.input)
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 50, in load_tf_models_file
return Unpickler(fp).load()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 45, in find_class
return super(Unpickler, self).find_class(module, name)
ModuleNotFoundError: No module named 'torch_utils'

My environment is cu101+torch1.7.1+tensorboard2.3.0 Please remind me if there is any information that is not enough, thanks a lot!

have you solve the problem? i got the same question.

[adriansahlman]

have you solve the problem? i got the same question.

The model you are trying to convert is not from the original tensorflow repo and thus will not work with this code

[a656418zz]

Hi, thank you very much for sharing. Due to the limitation of the GPU, I'm going to use "run_convert_from_tf.py" to transpose a 512x512 pre-trained model of stylegan2, which I got from https://catalog.ngc.nvidia.com/orgs/nvidia/teams/research/models/stylegan2 to download "stylegan2-ffhq-512x512.pkl", then I would like to try to conversion, but it failed, reporting the following error code.

Traceback (most recent call last):
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 373, in
main()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 350, in main
unpickled = load_tf_models_file(args.input)
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 50, in load_tf_models_file
return Unpickler(fp).load()
File "D:/PythonProject/StyleGAN/stylegan2_pytorch-adriansahlman/run_convert_from_tf.py", line 45, in find_class
return super(Unpickler, self).find_class(module, name)
ModuleNotFoundError: No module named 'torch_utils'

My environment is cu101+torch1.7.1+tensorboard2.3.0 Please remind me if there is any information that is not enough, thanks a lot!

have you solve the problem? i got the same question.

Unfortunately, because of adriansahlman‘s architecture, only the model he provides can be converted, the model of StyleGAN2-ada is newly generated and cannot be converted. But only the G model will not actually take up that much cuda memory, so I just generate 1080x1080 images and resize them

[a656418zz]

Hey!

It seems like the model you are trying to use is from https://github.com/NVlabs/stylegan2-ada-pytorch and not the original tensorflow repo that the conversion tool was designed for. I am sorry but these models are not supported by the current script as the repo I linked did not exist when I created this code!

Thanks for your help！