Open AK391 opened 3 years ago
Dear AK391,
Thank you for your interest in our project. Currently, we load the pretrained model from Stylegan2 which is trained using GPU. We can not even do Gs.print_layers() without GPU. Maybe try our colab which utilizes google gpu.
We will release a pure pythorch version in the future, which may run properly with just CPU.
Best Wishes,
Zongze
@betterze could anycost gan help with this https://github.com/mit-han-lab/anycost-gan
Dear AK391,
Thank you for your interest in our project. Currently, we load the pretrained model from Stylegan2 which is trained using GPU. We can not even do Gs.print_layers() without GPU. Maybe try our colab which utilizes google gpu.
We will release a pure pythorch version in the future, which may run properly with just CPU.
Best Wishes,
Zongze
Just a little question, why can not the tf-version code run on CPU?
Currently, we load the pretrained model from Stylegan2 which is trained using GPU.
I'm fresh in this field. Can't we load it using CPU and its memory?
I change the tf.configProto
to unuse GPU, but the code under global_directions
runs into error, which says:
tensorflow.python.framework.errors_impl.InvalidArgumentError: No OpKernel was registered to support Op 'FusedBiasAct' used by node G_synthesis_1/4x4/Conv/FusedBiasAct (defined at <string>:104) with these attrs: [T=DT_FLOAT, axis=1, grad=0, alpha=0, clamp=-1, act=1, gain=1]
Registered devices: [CPU, XLA_CPU, XLA_GPU]
Registered kernels:
device='GPU'; T in [DT_HALF]
device='GPU'; T in [DT_FLOAT]
[[G_synthesis_1/4x4/Conv/FusedBiasAct]]
Errors may have originated from an input operation.
Input Source operations connected to node G_synthesis_1/4x4/Conv/FusedBiasAct:
G_synthesis_1/4x4/Conv/mul_2 (defined at <string>:67)
G_synthesis_1/4x4/Conv/Const (defined at ./global_directions/dnnlib/tflib/ops/fused_bias_act.py:115)
G_synthesis_1/4x4/Conv/MatMul (defined at <string>:46)
A similar quesion: https://discuss.tensorflow.org/t/tensorflow-inference-on-cpu-after-training-model-on-gpu-with-amp-and-xla/12731
is it possible to run on cpu for inference, specifically the gui shown in the youtube video