Nvidia created 3D StyleganEG3D Efficient GeometryAware 3D Generate Adversarial Networks
- Jun 10
EG3D is an unstable engineering miracle. Resigners took the Stylegan 2 backbon as a basis, and inserted a hybrid neural rendering into the middle of the column.
NVIDIA created the new Tri-Plane-Bassed 3D Gan Framework. Reserves showed that Nerf in their case is slow, and the stakes are not very scaled from memory with an increase in permission.
After neural rendering in 128x128, the cascade of super-resolved in 512x512 follows. Then Low-Res once again appears interpolation and after concatenation with a mid-rem with a picture, a single tensor is fed to a double discriminator.
Stylegan’y are interesting in that they are capable of relatively small compounds (here 8 Tesla V100 GPUS and 8.5 days) build a realistic generative model in one domain.
EG3D allows you to immediately generate high -quality 3D geometry, which the network memorizes on its own!
How wonderful the future of 3D graphs!
Github:
Project:
Video:
Paper:
Documentation:
#machinelearning #artificialintelligence #ai #datascience #programming #Deechnology #Deeuplearning #bigData #bigdata