Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Novel Min-Max Reformulations of Linear Inverse Problems
Authors: Mohammed Rayyan Sheriff, Debasish Chatterjee
JMLR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We implement BPDN based image denoising on two images by solving (26) for all non overlapping patches of size 8 ˆ 8. Figure 1 shows the denoising results for the standard cameraman image, which is of size 256ˆ256. From left to right, we have the original image, noisy image and the recovered image in order. The noisy image is obtained by adding a mean zero Gaussian noise of standard deviation 0.0065 using imnoise function in MATLAB resulting in a PSNR (Peak Signal to Noise Ratio) of 22.0741d B. To recover the image, BPDN was solved with ϵ 0.3 for every non-overlapping 8ˆ8 patch using Algorithm 1 to compute the saddle point of the equivalent constrained min-max problem. |
| Researcher Affiliation | Academia | Mohammed Rayyan Sheriff EMAIL Debasish Chatterjee EMAIL Systems and Control Engineering IIT Bombay Mumbai 400076, India |
| Pseudocode | Yes | Algorithm 1: Projected gradient descent algorithm for constrained min-max problem (19). ... Algorithm 2: OGDA algorithm for unconstrained min-max problem (24) when δ 0. ... Algorithm 3: Projected gradient descent algorithm for constrained min-max problem (19). |
| Open Source Code | No | The paper does not provide an explicit statement about the release of source code for the methodology described, nor does it include a link to a code repository. The license information provided is for the paper itself, not for associated code. |
| Open Datasets | Yes | We implement BPDN based image denoising on two images by solving (26) for all non overlapping patches of size 8 ˆ 8. Figure 1 shows the denoising results for the standard cameraman image, which is of size 256ˆ256. ... Similarly, in Figure 2 the results for denoising the flower image are shown... |
| Dataset Splits | No | The paper mentions processing images using "non overlapping patches of size 8 ˆ 8" but does not specify any training, validation, or test dataset splits for the images used in the experiments. |
| Hardware Specification | No | The paper mentions that "The noisy image is obtained by adding a mean zero Gaussian noise of standard deviation 0.0065 using imnoise function in MATLAB resulting in a PSNR (Peak Signal to Noise Ratio) of 22.0741d B." However, it does not provide any specific details about the hardware (CPU, GPU, etc.) used to run the experiments or algorithms. |
| Software Dependencies | No | The paper mentions using "imnoise function in MATLAB" for generating noisy images. While MATLAB is a software, a specific version number is not provided, and no other software dependencies with version numbers are listed. |
| Experiment Setup | Yes | To recover the image, BPDN was solved with ϵ 0.3 for every non-overlapping 8ˆ8 patch using Algorithm 1... In Algorithm 1, we choose the step size sequence αt 5{p20 tq, ϵ 0.385 and for initialisation, h0 φ pxq φ pxq 1 and f0 x... Similarly, to compute a saddle point of the unconstrained min-max problem (28) via Algorithm 2, we use the step size sequence αt 2.5{p10 tq, ϵ 0.385 and for initialisation, f0 φ pxq and λ0 x. |