转载自: https://www.notion.so/Progressive-Growing-StyleGAN-StyleGAN2-83859f6f47ba4a508375e4e4cf0efd05
教学视频: https://www.youtube.com/playlist?list=PLl6jkZVaP8zrWGBRTmcy_HeVfDq27oXML
参考文献:
- Progressive Growing of GANs for Improved Quality, Stability, and Variation
- A Style-Based Generator Architecture for Generative Adversarial Networks
- Analyzing and Improving the Image Quality of StyleGAN
- https://nvlabs.github.io/stylegan2/versions.html
Progressive Growing GAN
Picture: Two imaginary celebrities that were dreamed up by a random number generator.
Result Video
https://www.youtube.com/watch?v=G06dEcZ-QTg
Main idea: Progressive Growth
The idea of multi resolution, multi-scale, pyramid, progressive itself is straightforward and not new.
For example LAPGAN back from 2015, SGAN 2016,
LAPGAN sampling
LAPGAN training
PGGAN
StyleGAN
AdaIN
In StyleGAN, y y y is computed from w w w instead of an example image.
Perceptual path length
Intend to measure how smooth the interpolation of latent vectors is.
The idea is to want the path length to be short in some perceptual space.
Truncation trick
StyleGAN2
https://www.youtube.com/watch?v=c-NJtV9Jvp0
Droplet-like artifacts→ alternative normalization
- modulation: scaling the convolution weights instead of the features: w i j k ′ = s i ⋅ w i j k w'_{ijk}=s_i\cdot w_{ijk} wijk′=si⋅wijk
- by assuming the input activations are i.i.d with unit standard deviation, after modulation an c convolution, the output have standard deviation of σ j = ∑ i , j w i j k ′ 2 \sigma_j=\sqrt{\sum_{i,j} {w'_{ijk} }^2} σj=∑i,jwijk′2
- so, demodulation: w i j k ′ ′ = w i j k ′ ∑ i , k w i j k ′ 2 + ϵ w''_{ijk} =\frac{w'_{ijk}}{\sqrt{\sum_{i,k}{w'_{ijk}}^2+\epsilon}} wijk′′=∑i,kwijk′2+ϵwijk′
- The style process can be baked into a convolution
- a.k.a weight normalization
- removes the artifacts, same FID.
Image quality and generator smoothness
The regularization
Can compute efficiently with
- a a a is set dynamically during optimization as the long running exponential moving average of the lengths ∥ J w y ∥ 2 \Vert \mathbf{J_wy}\Vert_2 ∥Jwy∥2.
- closely related to Jacobian clamping regularization.
- Better FID than spectral normalization
Progressive growing revisited