Learning optimized MAP estimates in continuously-valued MRF models
read more
Citations
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising
FFDNet: Toward a Fast and Flexible Solution for CNN based Image Denoising
Shrinkage Fields for Effective Image Restoration
Solving inverse problems using data-driven models
References
Image quality assessment: from error visibility to structural similarity
A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics
Training products of experts by minimizing contrastive divergence
Choosing Multiple Parameters for Support Vector Machines
Image and depth from a conventional camera with a coded aperture
Related Papers (5)
Frequently Asked Questions (11)
Q2. What is the method to learn the parameters of a continuous-state MRF model?
Using a chain-structured model makes it possible to compute the Hessian of the CRF’s density function, thus enabling the method to learn hyper-parameters.
Q3. What is the common method of ensuring positive weighting coefficients?
For instance, it is common to use an exponential function to ensure positive weighting coefficients, such as w(θ) = eθ for a scalar θ, as employed in work like [12].
Q4. What is the key advantage of the Variational Mode Learning approach?
The key advantage of their approach over Variational Mode Learning, is that the result of the variational optimization must be recomputed every time the gradient of the loss function is recomputed.
Q5. How can the authors compute the gradient of a vector?
by computing ∂L(x*(θ),t) ∂θi T HE(x*)−1 rather than HE(x*)−1 ∂g ∂w ∂w ∂θi, only one call to the solver is necessary to compute gradient for all parameters in the vector θ.
Q6. How do you find the results with the highest PSNR?
Using this model, the best results, with the highest PSNR, are found by terminating the minimization of the negative loglikelihood of the model before reaching a local minimum.
Q7. How do the authors train the MRF model?
As will be explained in Section 2, the authors train the MRF model by optimizing its parameters so that the minimum energy solution of the model is as similar as possible to the ground-truth.
Q8. What is the argument for a probabilistic approach to learning the parameters of a continuousvalue?
While Section 2.1 explores this issue in more detail, their high-level argument is that if one’s goal is to use MAP estimates and evaluate the results using some criterion then optimizing MRF parameters using a probabilistic criterion, like maximum-likelihood parameter estimation, will not necessarily lead to the optimal system.
Q9. What is the gradient of the energy function E(x, y, )?
In this case the gradient of the energy function E(x, y;α, β) can then be written as∂E(x, y;α, β) ∂x = FT ρ′(u) + 2(x − y) (12)where ρ′(u) is a function that is applied elementwise to the vector u =
Q10. What is the strategy for maximizing the PSNR?
Returning to the example in the previous paragraph, if the goal is to maximize PSNR, maximizing a likelihood and then hoping that it leads to MAP estimates with good PSNR values is not the best strategy.
Q11. How can the authors learn the parameters of a hyperparameter?
the similarity of Equation 1 to solutions for learning hyper-parameters, such as [3, 1, 6, 2], suggests that it may be possible to optimize the MRF parameters θ using simpler gradient-based algorithms.