{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/neural-proximal-gradient-descent-for","title":"Neural Proximal Gradient Descent for Compressive Imaging","arxiv_id":"1806.03963","date":"2018-06-01","proceeding":"NeurIPS 2018 12","authors":["Morteza Mardani","Qingyun Sun","Shreyas Vasawanala","Vardan Papyan","Hatef Monajemi","John Pauly","David Donoho"],"abstract":"Recovering high-resolution images from limited sensory data typically leads\nto a serious ill-posed inverse problem, demanding inversion algorithms that\neffectively capture the prior information. Learning a good inverse mapping from\ntraining data faces severe challenges, including: (i) scarcity of training\ndata; (ii) need for plausible reconstructions that are physically feasible;\n(iii) need for fast reconstruction, especially in real-time applications. We\ndevelop a successful system solving all these challenges, using as basic\narchitecture the recurrent application of proximal gradient algorithm. We learn\na proximal map that works well with real images based on residual networks.\nContraction of the resulting map is analyzed, and incoherence conditions are\ninvestigated that drive the convergence of the iterates. Extensive experiments\nare carried out under different settings: (a) reconstructing abdominal MRI of\npediatric patients from highly undersampled Fourier-space data and (b)\nsuperresolving natural face images. Our key findings include: 1. a recurrent\nResNet with a single residual block unrolled from an iterative algorithm yields\nan effective proximal which accurately reveals MR image details. 2. Our\narchitecture significantly outperforms conventional non-recurrent deep ResNets\nby 2dB SNR; it is also trained much more rapidly. 3. It outperforms\nstate-of-the-art compressed-sensing Wavelet-based methods by 4dB SNR, with 100x\nspeedups in reconstruction time.","url_abs":"http://arxiv.org/abs/1806.03963v1","url_pdf":"http://arxiv.org/pdf/1806.03963v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"neural-proximal-gradient-descent-for","repo_url":"https://github.com/MortezaMardani/Neural-PGD","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"compressed-sensing","task_name":"compressed sensing"}],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.03963","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}