Adversarial Generation of Continuous Images
Published:
neural implicit representation and GAN
- using a hyper network
- train a shared weight prior and a offset for each layer
- factorized multiplicative modulation
- weight of each frequency predicted by latent variable
- better latent space: keypoint estimation from latent code(disentanglement)
- multiscale: append position from different resolution
- another way to progressively grow network