{"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/analyzing-perception-distortion-tradeoff","title":"Analyzing Perception-Distortion Tradeoff using Enhanced Perceptual Super-resolution Network","arxiv_id":"1811.00344","date":"2018-11-01","proceeding":null,"authors":["Subeesh Vasu","Nimisha Thekke Madam","Rajagopalan A. N"],"abstract":"Convolutional neural network (CNN) based methods have recently achieved great\nsuccess for image super-resolution (SR). However, most deep CNN based SR models\nattempt to improve distortion measures (e.g. PSNR, SSIM, IFC, VIF) while\nresulting in poor quantified perceptual quality (e.g. human opinion score,\nno-reference quality measures such as NIQE). Few works have attempted to\nimprove the perceptual quality at the cost of performance reduction in\ndistortion measures. A very recent study has revealed that distortion and\nperceptual quality are at odds with each other and there is always a trade-off\nbetween the two. Often the restoration algorithms that are superior in terms of\nperceptual quality, are inferior in terms of distortion measures. Our work\nattempts to analyze the trade-off between distortion and perceptual quality for\nthe problem of single image SR. To this end, we use the well-known SR\narchitecture-enhanced deep super-resolution (EDSR) network and show that it can\nbe adapted to achieve better perceptual quality for a specific range of the\ndistortion measure. While the original network of EDSR was trained to minimize\nthe error defined based on per-pixel accuracy alone, we train our network using\na generative adversarial network framework with EDSR as the generator module.\nOur proposed network, called enhanced perceptual super-resolution network\n(EPSR), is trained with a combination of mean squared error loss, perceptual\nloss, and adversarial loss. Our experiments reveal that EPSR achieves the\nstate-of-the-art trade-off between distortion and perceptual quality while the\nexisting methods perform well in either of these measures alone.","url_abs":"http://arxiv.org/abs/1811.00344v2","url_pdf":"http://arxiv.org/pdf/1811.00344v2.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":"analyzing-perception-distortion-tradeoff","repo_url":"https://github.com/subeeshvasu/2018_subeesh_epsr_eccvw","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":null,"task_name":"Generative Adversarial Network"},{"task_slug":"image-super-resolution","task_name":"Image Super-Resolution"},{"task_slug":"ssim","task_name":"SSIM"},{"task_slug":"super-resolution","task_name":"Super-Resolution"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.00344","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}