{"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/deep-convolution-networks-for-compression","title":"Deep Convolution Networks for Compression Artifacts Reduction","arxiv_id":"1608.02778","date":"2016-08-09","proceeding":null,"authors":["Ke Yu","Chao Dong","Chen Change Loy","Xiaoou Tang"],"abstract":"Lossy compression introduces complex compression artifacts, particularly\nblocking artifacts, ringing effects and blurring. Existing algorithms either\nfocus on removing blocking artifacts and produce blurred output, or restore\nsharpened images that are accompanied with ringing effects. Inspired by the\nsuccess of deep convolutional networks (DCN) on superresolution, we formulate a\ncompact and efficient network for seamless attenuation of different compression\nartifacts. To meet the speed requirement of real-world applications, we further\naccelerate the proposed baseline model by layer decomposition and joint use of\nlarge-stride convolutional and deconvolutional layers. This also leads to a\nmore general CNN framework that has a close relationship with the conventional\nMulti-Layer Perceptron (MLP). Finally, the modified network achieves a speed up\nof 7.5 times with almost no performance loss compared to the baseline model. We\nalso demonstrate that a deeper model can be effectively trained with features\nlearned in a shallow network. Following a similar \"easy to hard\" idea, we\nsystematically investigate three practical transfer settings and show the\neffectiveness of transfer learning in low-level vision problems. Our method\nshows superior performance than the state-of-the-art methods both on benchmark\ndatasets and a real-world use case.","url_abs":"http://arxiv.org/abs/1608.02778v1","url_pdf":"http://arxiv.org/pdf/1608.02778v1.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":"deep-convolution-networks-for-compression","repo_url":"https://github.com/ankitf/artifact_reduction_jpeg","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"deep-convolution-networks-for-compression","repo_url":"https://github.com/vinayak19th/ARCNN-keras","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"blocking","task_name":"Blocking"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1608.02778","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}