Learning optimized MAP estimates in continuously-valued MRF models

By Kegan G. G. Samuel, Marshall F. Tappen. 2009

113 citations recorded by the source corpus.

Subfields: Biophysics, Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

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

In Biophysics

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

On Differentiating Parameterized Argmin and Argmax Problems with Application to Bi-level Optimization (2016, methodology): “…vision where they have been applied to parameter and hyper-parameter learning [Do et al., 2007, Domke, 2012, Klatzer and Pock, 2015], image denoising [Samuel and Tappen, 2009, Ochs et al., 2015], and most recently, video activity recognition [Fernando and Gould, 2016, Fernando et al., 2016].”
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising (2016, background): “Some of the other related work can be found in [20], [21], [22], [23], [24], [25].”
Parameter Learning and Convergent Inference for Dense Random Fields (2013, methodology): “We take a different approach and directly minimize a loss function on the approximate distribution that is used by the inference algorithm (Tappen, 2007; Samuel & Tappen, 2009).”
Shrinkage Fields for Effective Image Restoration (2014, methodology): “Gaussian CRFs and their associated loss-based training have ﬁrst been introduced by Tappen [27].”
Power Watershed: A Unifying Graph-Based Optimization Framework (2011, background): “Beyond Gaussian MRFs, anisotropic diffusion has been interpreted as a continuous-valued MRF [61], [51] and MRFs requiring continuous-valued estimations have appeared in both early work on computer vision [55], [37], [35], [16] and also recently [56], [74], [75].”
Deep Unfolding Network for Image Super-Resolution (2020, background): “The early deep unfolding meth-ods can be traced back to [4, 48, 54] where a compact MAP inference based on gradient descent algorithm is proposed for image denoising.”
FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising (2017, unknown): “Sun and Tappen [23] proposed a novel nonlocal range MRF (NLR-MRF) framework, and employed the gradient-based discriminative learning method to train the model.”
Dataset Condensation with Gradient Matching (2020, background): “(3) involves a nested loop optimization and solving the inner loop for θS(S) at each iteration to recover the gradients for S which requires a computationally expensive procedure – unrolling the recursive computation graph for S over multiple optimization steps for θ (see [34, 35]).”
Generic Methods for Optimization-Based Modeling (2012, background): “(1) Such mappings are common in structured prediction tasks such as denoising (Samuel and Tappen, 2009, Sun and Tappen, 2011, Barbu, 2009) or image labeling (Tappen et al., 2008, 2007).”
Differentiation of Blackbox Combinatorial Solvers (2019, methodology): “However, the function f ( y(w) ) is a result of a minimization process and it is known that for smooth spaces Y there are techniques for such “differentiation through argmin” [2, 3, 11, 13, 38, 40].”
End-to-End Learning for Structured Prediction Energy Networks (2017, unknown): “The implicit function theorem offers an alternative framework for training energy-based predictors (Foo et al., 2008; Samuel & Tappen, 2009).”
A Bilevel Optimization Approach for Parameter Learning in Variational Models (2013, methodology): “We do not exactly know the reason for the improved performance of our simpler model, but possibly our Newton algorithms are distinctly more accurate in approximating (locally) optimal solutions than the gradient descent methods that are used in [27].”
Reviving and Improving Recurrent Back-Propagation (2018, methodology): “A similar implicit differentiation technique is used in (Samuel & Tappen, 2009) to optimize the maximum a posterior (MAP) solution of continuous MRFs, since the MAP solution can be regarded as the steady state of the inference process.”
Deep Declarative Networks (2019, unknown): “Samuel and Tappen [48] propose learning param-”