{"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/pieapp-perceptual-image-error-assessment","title":"PieAPP: Perceptual Image-Error Assessment through Pairwise Preference","arxiv_id":"1806.02067","date":"2018-06-06","proceeding":"CVPR 2018 6","authors":["Ekta Prashnani","Hong Cai","Yasamin Mostofi","Pradeep Sen"],"abstract":"The ability to estimate the perceptual error between images is an important\nproblem in computer vision with many applications. Although it has been studied\nextensively, however, no method currently exists that can robustly predict\nvisual differences like humans. Some previous approaches used hand-coded\nmodels, but they fail to model the complexity of the human visual system.\nOthers used machine learning to train models on human-labeled datasets, but\ncreating large, high-quality datasets is difficult because people are unable to\nassign consistent error labels to distorted images. In this paper, we present a\nnew learning-based method that is the first to predict perceptual image error\nlike human observers. Since it is much easier for people to compare two given\nimages and identify the one more similar to a reference than to assign quality\nscores to each, we propose a new, large-scale dataset labeled with the\nprobability that humans will prefer one image over another. We then train a\ndeep-learning model using a novel, pairwise-learning framework to predict the\npreference of one distorted image over the other. Our key observation is that\nour trained network can then be used separately with only one distorted image\nand a reference to predict its perceptual error, without ever being trained on\nexplicit human perceptual-error labels. The perceptual error estimated by our\nnew metric, PieAPP, is well-correlated with human opinion. Furthermore, it\nsignificantly outperforms existing algorithms, beating the state-of-the-art by\nalmost 3x on our test set in terms of binary error rate, while also\ngeneralizing to new kinds of distortions, unlike previous learning-based\nmethods.","url_abs":"http://arxiv.org/abs/1806.02067v1","url_pdf":"http://arxiv.org/pdf/1806.02067v1.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":"pieapp-perceptual-image-error-assessment","repo_url":"https://github.com/prashnani/PerceptualImageError","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"video-quality-assessment","task_name":"Video Quality Assessment"}],"methods":[],"datasets_introduced":[{"slug":"pieapp-dataset","name":"PieAPP dataset","full_name":"PieAPP dataset"}],"methods_introduced":[],"results":[{"leaderboard":"/sota/video-quality-assessment-on-msu-sr-qa-dataset","task":"Video Quality Assessment","dataset":"MSU SR-QA Dataset","model":"PieAPP","rank_in_archive_order":1,"of":60,"metrics":{"KLCC":"0.61945","PLCC":"0.75743","SROCC":"0.75215","Type":"FR"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1806.02067","atlas_url":"https://app.syntology.ai/?focus=1806.02067","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}