- System Standards
- Knowledge time
- One another Linux and you can Window was supported, but i strongly recommend Linux having results and you may being compatible reasons.
- 64-portion Python step three.6 construction. We recommend Anaconda3 having numpy step 1.fourteen.step 3 otherwise latest.
- TensorFlow step 1.10.0 otherwise latest having GPU help.
- One or more large-prevent NVIDIA GPUs which have about 11GB regarding DRAM. We recommend NVIDIA DGX-step one with 8 Tesla V100 GPUs.
- NVIDIA driver or latest, CUDA toolkit 9.0 otherwise new, cuDNN 7.3.step one otherwise latest.
Degree go out
Below there is certainly NVIDIA’s reported questioned degree times having default configuration of the program (for sale in the latest stylegan repository) on good Tesla V100 GPU into FFHQ dataset (found in the fresh new stylegan data source).
Behind the scenes
It developed the StyleGAN. To know a little more about the subsequent approach, We have considering certain resources and you can to the level causes below.
Generative Adversarial Community
Generative Adversarial Sites first-made the brand new series into the 2014 due to the fact an enthusiastic expansion out of generative patterns via a keen adversarial processes where we as well train a few patterns: