"Natural Images, Gaussian Mixtures and Dead Leaves"

By Daniel Zoran, Yair Weiss. 2012

113 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computer Vision and Pattern Recognition

Citations in the corpus

Neural Autoregressive Distribution Estimation (2016, methodology): “e In this section, we compare the statistical performance of RNADE to mixtures of Gaussians (MoG) and factor analyzers (MFA), which are surprisingly strong baselines in some tasks (Tang et al., 2012; Zoran and Weiss, 2012). 7.3.1 Low-dimensional data We start by considering three UCI datasets (Bache”
Generative Image Modeling Using Spatial LSTMs (2015, neutral): “s. We 4 Model 63 dim. [nat] 64 dim. [bit/px] 1 dim. [bit/px] RNADE [44] 152.1 3.346 - RNADE, 1 hl [45] 143.2 3.146 - RNADE, 6 hl [45] 155.2 3.416 - EoRNADE, 6 layers [45] 157.0 3.457 - GMM, 200 comp. [47, 50] 153.7 3.360 - STM, 200 comp. [46] 155.3 3.418 - Deep GMM, 3 layers [47] 156.2 3.439 - MCGSM”
Video (language) modeling: a baseline for generative models of natural videos (2014, background): “Although some recent works have shown some progress on smaller resolution images (Goodfellow et al., 2014; Zoran & Weiss, 2012; Ranzato et al., 2013; Theis et al., 2011) using generative models, it is unclear how these models can scale to handle full resolution inputs, due to the curse of…”
A Deep and Tractable Density Estimator (2013, methodology): “We compare the performance of RNADEs with different number of hidden layers trained with our procedure against a one-hidden layer RNADE trained for a ﬁxed ordering (Uria et al., 2013), and with mixtures of Gaussians, which remain the state of the art in this problem (Zoran & Weiss, 2012).”
Density Modeling of Images using a Generalized Normalization Transformation (2015, neutral): “…this induced density model more thoroughly to other model forms that have been proposed in the literature (e.g., ﬁnite mixtures of Gaussians or GSMs (Guerrero-Colón et al., 2008; Lyu & Simoncelli, 2009a; Zoran & Weiss, 2012; Theis et al., 2012), and sparse factorization (Culpepper et al., 2011)).”
Crisp Boundary Detection Using Pointwise Mutual Information (2014, background): “The rarity of boundary patches has been studied in the literature before, e.g. [23].”
The student-t mixture as a natural image patch prior with application to image compression (2014, methodology): “Apart from being simple and efficient, GMMs are easily interpretable methods which allow us to learn more about the nature of images (Zoran and Weiss, 2012).”
Connectionist multivariate density-estimation and its application to speech synthesis (2016, methodology): “Model Test log-likelihood MoG K=200 (Zoran and Weiss, 2012)a 152.”
On the Expressive Power of Deep Learning: A Tensor Analysis (2015, unknown): “For example, it was shown in Zoran and Weiss (2012) that natural image patches of size 2×2, 4×4, 8×8 or 16×16, can essentially be modeled by GMMs with 64 components (adding more components barely improved the log-likelihood).”
Patch Group Based Nonlocal Self-Similarity Prior Learning for Image Denoising (2015, background): “By viewing image patches as samples of a multivariate variable vector and considering that natural images are non-Gaussian, Zoran and Weiss [26, 27] and Yu et al.”