Revisiting Bayesian Blind Deconvolution

Authors: David Wipf, Haichao Zhang

JMLR 2014 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Section 6. Experimental Results describes the evaluation of algorithms using benchmark test data and performance metrics. Figure 3 shows cumulative histograms of error ratios for comparison.
Researcher Affiliation Collaboration David Wipf is affiliated with Microsoft Research, and Haichao Zhang is affiliated with Northwestern Polytechnical University, indicating a collaboration between industry and academia.
Pseudocode Yes The paper includes 'Algorithm 1 VB Blind Deblurring' and 'Algorithm 2 VB Blind Deblurring with Jeffreys Prior and Learned λ (VB+)', which are clearly structured algorithm blocks.
Open Source Code No The paper does not contain an explicit statement that the authors provide open-source code for the methodology described in this paper. It mentions using 'a script provided directly from the first author s website' for VB-Babacan (a third-party algorithm), but not for their own work.
Open Datasets Yes The paper states: 'we reproduce the experiments from Levin et al. (2011a) using the useful benchmark test data from Levin et al. (2009).' Footnote 17 provides a direct link: 'This data is available online at http://www.wisdom.weizmann.ac.il/~levina/papers/Levin_Etal_CVPR09Data.rar'.
Dataset Splits No The paper uses a benchmark test dataset consisting of 32 blurry images for evaluation. However, it does not specify any training, validation, or test splits for these images, as the experiments focus on evaluating algorithms on a fixed test set rather than training new models with partitioned data.
Hardware Specification No The paper does not provide specific details about the hardware used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes The paper provides specific experimental setup details, including: 'For this reduction schedule of λ we use β = 1.15 in Algorithm 1' (Section 3.6), and 'We set d = n * 10^-4 for all simulations' for Algorithm 2 (Section 5). It also mentions initialization of blur kernel k and noise level λ in Algorithm 1 and 2 inputs.