search

dvschultz / stylegan2-ada-pytorch

StyleGAN2-ADA - Official PyTorch implementation
https://arxiv.org/abs/2006.06676
Other
244 stars 121 forks source link

Unable to generate single images after training with the notebook #44

Open chasetyoung opened 1 year ago

chasetyoung commented 1 year ago

Unable to generate single images after training with the notebook.

After training a model, the notebook is unable to generate image results. When the following command is run, the following error occurs- !python generate.py --outdir=/content/out/images/ --trunc=0.8 --seeds=0 --network=/content/drive/MyDrive/ladiescrop-network-snapshot-012885.pkl

generate.py:59: SyntaxWarning: "is not" with a literal. Did you mean "!="? elif(len(seeds) is not 3): Traceback (most recent call last): File "generate.py", line 25, in from opensimplex import OpenSimplex File "/usr/local/lib/python3.8/dist-packages/opensimplex/init.py", line 4, in from .api import * File "/usr/local/lib/python3.8/dist-packages/opensimplex/api.py", line 2, in from .internals import _init, _noise2, _noise3, _noise4, _noise2a, _noise3a, _noise4a File "/usr/local/lib/python3.8/dist-packages/opensimplex/internals.py", line 9, in from numba import njit, prange File "/usr/local/lib/python3.8/dist-packages/numba/init.py", line 42, in from numba.np.ufunc import (vectorize, guvectorize, threading_layer, File "/usr/local/lib/python3.8/dist-packages/numba/np/ufunc/init.py", line 3, in from numba.np.ufunc.decorators import Vectorize, GUVectorize, vectorize, guvectorize File "/usr/local/lib/python3.8/dist-packages/numba/np/ufunc/decorators.py", line 3, in from numba.np.ufunc import _internal SystemError: initialization of _internal failed without raising an exception

daanrongen commented 1 year ago

I'm having the same problem. It seems like an issue with the opensimplex import

daanrongen commented 1 year ago

So the issue is within the interpolation logic to generate video outputs (interpolating from seed to seed). opensimplex is used in the OSN() class that is used to predict noise in a noiseloop. If you comment all these out (along with the if else logic from generate_images(), it works. But you'll only be able to use interpolation=linear in generate.py.