Block Coordinate Regularization by Denoising
Authors: Yu Sun, Jiaming Liu, Ulugbek Kamilov
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We numerically validate our method using several denoiser priors, including those based on convolutional neural network (CNN) denoisers. Additionally, as we shall see, the overall computational complexity of BC-RED can sometimes be lower than corresponding methods operating on the full vector. This behavior is consistent with the traditional coordinate descent methods that can outperform their full gradient counterparts by being able to better reuse local updates and take larger steps [25 29]. We present two theoretical results related to BC-RED. We first theoretically characterize the convergence of the algorithm and discuss its relationship to the traditional proximal optimization. Our analysis complements and extends recent theoretical results for RED-based estimation methods. We numerically validate our method using several denoiser priors, including those based on convolutional neural network (CNN) denoisers. |
| Researcher Affiliation | Academia | Yu Sun Washington University in St. Louis sun.yu@wustl.edu Jiaming Liu Washington University in St. Louis jiaming.liu@wustl.edu Ulugbek S. Kamilov Washington University in St. Louis kamilov@wustl.edu |
| Pseudocode | Yes | Algorithm 1 Block Coordinate Regularization by Denoising (BC-RED) |
| Open Source Code | Yes | The code for our implementation of BC-RED is available through the following link1. 1https://github.com/wustl-cig/bcred |
| Open Datasets | Yes | The images used correspond to 10 images randomly selected from the NYU fast MRI dataset [59], resized to be 160 160 pixels. [59] Zbontar et al., fast MRI: An open dataset and benchmarks for accelerated MRI, 2018, ar Xiv:1811.08839. |
| Dataset Splits | No | The paper mentions selecting denoisers for highest SNR, implying a validation process, but does not provide specific dataset split percentages or sample counts for training, validation, or test sets. |
| Hardware Specification | Yes | Acknowledgments: This material is based upon work supported in part by NSF award CCF-1813910 and by NVIDIA Corporation with the donation of the Titan Xp GPU for research. |
| Software Dependencies | No | The paper mentions software like 'Dn CNN' but does not specify version numbers for any key software components or libraries. |
| Experiment Setup | Yes | In all simulations, we set the measurement ratio to be approximately m/n = 0.5 with AWGN corresponding to input signal-to-noise ratio (SNR) of 30 d B and 40 d B. The images used correspond to 10 images randomly selected from the NYU fast MRI dataset [59], resized to be 160 160 pixels. BC-RED is set to work with 16 blocks, each of size 40 40 pixels. ... We train Dn CNN for the removal of AWGN at four noise levels corresponding to σ {5, 10, 15, 20}. |