{"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/super-resolution-with-deep-convolutional","title":"Super-Resolution with Deep Convolutional Sufficient Statistics","arxiv_id":"1511.05666","date":"2015-11-18","proceeding":null,"authors":["Joan Bruna","Pablo Sprechmann","Yann Lecun"],"abstract":"Inverse problems in image and audio, and super-resolution in particular, can\nbe seen as high-dimensional structured prediction problems, where the goal is\nto characterize the conditional distribution of a high-resolution output given\nits low-resolution corrupted observation. When the scaling ratio is small,\npoint estimates achieve impressive performance, but soon they suffer from the\nregression-to-the-mean problem, result of their inability to capture the\nmulti-modality of this conditional distribution. Modeling high-dimensional\nimage and audio distributions is a hard task, requiring both the ability to\nmodel complex geometrical structures and textured regions. In this paper, we\npropose to use as conditional model a Gibbs distribution, where its sufficient\nstatistics are given by deep convolutional neural networks. The features\ncomputed by the network are stable to local deformation, and have reduced\nvariance when the input is a stationary texture. These properties imply that\nthe resulting sufficient statistics minimize the uncertainty of the target\nsignals given the degraded observations, while being highly informative. The\nfilters of the CNN are initialized by multiscale complex wavelets, and then we\npropose an algorithm to fine-tune them by estimating the gradient of the\nconditional log-likelihood, which bears some similarities with Generative\nAdversarial Networks. We evaluate experimentally the proposed approach in the\nimage super-resolution task, but the approach is general and could be used in\nother challenging ill-posed problems such as audio bandwidth extension.","url_abs":"http://arxiv.org/abs/1511.05666v4","url_pdf":"http://arxiv.org/pdf/1511.05666v4.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":"super-resolution-with-deep-convolutional","repo_url":"https://github.com/Adk2001tech/GAN-Image-Super-Resolution","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"bandwidth-extension","task_name":"Bandwidth Extension"},{"task_slug":"image-super-resolution","task_name":"Image Super-Resolution"},{"task_slug":"structured-prediction","task_name":"Structured Prediction"},{"task_slug":"super-resolution","task_name":"Super-Resolution"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1511.05666","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}