Image Modeling with Deep Convolutional Gaussian Mixture Models

被引:0
|
作者
Gepperth, Alexander [1 ]
Pfuelb, Benedikt [1 ]
机构
[1] Fulda Univ Appl Sci, Fulda, Germany
关键词
Deep Learning; Gaussian Mixture Models; Deep Convolutional Gaussian Mixture Models; Stochastic Gradient Descent;
D O I
10.1109/IJCNN52387.2021.9533745
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In this conceptual work, we present Deep Convolutional Gaussian Mixture Models (DCGMMs): a new formulation of deep hierarchical Gaussian Mixture Models (GMMs) that is particularly suitable for describing and generating images. Vanilla (i.e., flat) GMMs require a very large number of components to describe images well, leading to long training times and memory issues. DCGMMs avoid this by a stacked architecture of multiple GMM layers, linked by convolution and pooling operations. This allows to exploit the compositionality of images in a similar way as deep CNNs do. DCGMMs can be trained end-to-end by Stochastic Gradient Descent. This sets them apart from vanilla GMMs which are trained by Expectation-Maximization, requiring a prior k-means initialization which is infeasible in a layered structure. For generating sharp images with DCGMMs, we introduce a new gradient-based technique for sampling through non-invertible operations like convolution and pooling. Based on the MNIST and FashionMNIST datasets, we validate the DCGMMs model by demonstrating its superiority over flat GMMs for clustering, sampling and outlier detection.
引用
收藏
页数:9
相关论文
共 50 条
  • [1] Deep Gaussian mixture models
    Cinzia Viroli
    Geoffrey J. McLachlan
    [J]. Statistics and Computing, 2019, 29 : 43 - 51
  • [2] Deep Gaussian mixture models
    Viroli, Cinzia
    McLachlan, Geoffrey J.
    [J]. STATISTICS AND COMPUTING, 2019, 29 (01) : 43 - 51
  • [3] Convolutional Gaussian Mixture Models with Application to Compressive Sensing
    Wang, Ren
    Liao, Xuejun
    Guo, Jingbo
    [J]. 2018 IEEE STATISTICAL SIGNAL PROCESSING WORKSHOP (SSP), 2018, : 298 - 302
  • [4] Modeling Image Similarity by Gaussian Mixture Models and the Signature Quadratic Form Distance
    Beecks, Christian
    Ivanescu, Anca Maria
    Kirchhoff, Steffen
    Seidl, Thomas
    [J]. 2011 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), 2011, : 1754 - 1761
  • [5] Infrared image segmentation with Gaussian mixture modeling
    Lee, Dong-Su
    Yeom, Seokwon
    [J]. INFRARED IMAGING SYSTEMS: DESIGN, ANALYSIS, MODELING, AND TESTING XXIII, 2012, 8355
  • [6] Image Fusion Using Gaussian Mixture Models
    Heaps, Katie
    Koslosky, Josh
    Sidle, Glenn
    Levine, Stacey
    [J]. PROCEEDINGS OF THE BRITISH MACHINE VISION CONFERENCE 2013, 2013,
  • [7] Nonlinear prediction for Gaussian mixture image models
    Zhang, J
    Ma, DH
    [J]. VISUAL COMMUNICATIONS AND IMAGE PROCESSING 2001, 2001, 4310 : 99 - 110
  • [8] Multiple Gaussian Mixture Models for Image Registration
    Ye, Peng
    Liu, Fang
    Zhao, Zhiyong
    [J]. IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS, 2014, E97D (07) : 1927 - 1929
  • [9] Nonlinear prediction for Gaussian mixture image models
    Zhang, J
    Ma, DH
    [J]. IEEE TRANSACTIONS ON IMAGE PROCESSING, 2004, 13 (06) : 836 - 847
  • [10] Gaussian Mixture Modeling with Gaussian Process Latent Variable Models
    Nickisch, Hannes
    Rasmussen, Carl Edward
    [J]. PATTERN RECOGNITION, 2010, 6376 : 272 - 282