Sampling Generative Networks Notes on a Few Effective Techniques

By Tom White. 2016

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

Citations in the corpus

A Style-Based Generator Architecture for Generative Adversarial Networks (2018, background): “Here slerp denotes spherical interpolation [52], which is the most appropriate way of interpolating in our normalized input latent space [56].”
GAN Inversion: A Survey (2021, background): “Typically, slerp(·) in Z denotes spherical interpolation [63], which is the most appropriate way of interpolating in the normalized input latent space [64].”
Pros and Cons of GAN Evaluation Measures (2018, background): “White [117] suggests that replacing linear interpolation with spherical linear interpolation prevents diverging from a model’s prior distribution and produces sharper samples.”
COCO-GAN: Generation by Parts via Conditional Coordinating (2019, methodology): “With any given interpolation point z in the slerp-path [27] between z1 and z2, the generator uses the full spatial coordinate sequence C full to generate all corresponding patches.”
A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music (2018, methodology): “In practice we therefore use spherical interpolation (White, 2016) instead of Eq.”
An Introduction to Variational Autoencoders (2019, methodology): “Methods and VAE architectures for controlled image generation from attributes or text were studied in (Kingma et al., 2014; Yan et al., 2016; Mansimov et al., 2015; Brock et al., 2017; White, 2016).”
Latent Constraints: Learning to Generate Conditionally from Unconditional Generative Models (2017, background): “White (2016) propose adding “attribute vectors” to samples from p ( z ) as a simple and effective heuristic to perform transformations, which relies heavily on the linearity of the latent space.”
Distribution-Interpolation Trade off in Generative Models (2018, methodology): “As in Shoemake (1985); White (2016), the spherical linear interpolation is deﬁned as where Ω is the angle between vectors x 1 and x 2 .”
Semantic Interpolation in Implicit Models (2017, methodology): “We have already eluded to the proposal of White (2016), which uses the interpolation with θ := ∠ ( z 0 , z 1 ) , It is easy to check that if (cid:107) z 0 (cid:107) = (cid:107) z 1 (cid:107) = r , then this curve follows the great circle with radius r that connects z 0 and z 1 .”
Procedural Content Generation via Machine Learning (PCGML) (2017, background): “space image of blank-faced person from a smiling one is the smile vector) [102].”
Controlling generative models with continuous factors of variations (2020, background): “uctions signiﬁcantly. We also theoretically justify the difﬁculties to invert a generative model, compared to other optimization problems. In the context of vector space arithmetic in a latent space, White (2016) argues that replacing a linear interpolation by a spherical one allows to reduce the bl”
Learning a Latent Space of Multitrack Measures (2018, methodology): “…two measures x 0 and x 1 , we can interpolate between them by applying the encoder to obtain latent codes z 0 and z 1 , then for any 0 ≤ α ≤ 1 constructing z α using spherical linear interpolation [41], as most of the probability mass of the Gaussian prior lies very close to the unit hypersphere.”