{"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/photo-aesthetics-ranking-network-with","title":"Photo Aesthetics Ranking Network with Attributes and Content Adaptation","arxiv_id":"1606.01621","date":"2016-06-06","proceeding":null,"authors":["Shu Kong","Xiaohui Shen","Zhe Lin","Radomir Mech","Charless Fowlkes"],"abstract":"Real-world applications could benefit from the ability to automatically\ngenerate a fine-grained ranking of photo aesthetics. However, previous methods\nfor image aesthetics analysis have primarily focused on the coarse, binary\ncategorization of images into high- or low-aesthetic categories. In this work,\nwe propose to learn a deep convolutional neural network to rank photo\naesthetics in which the relative ranking of photo aesthetics are directly\nmodeled in the loss function. Our model incorporates joint learning of\nmeaningful photographic attributes and image content information which can help\nregularize the complicated photo aesthetics rating problem.\n  To train and analyze this model, we have assembled a new aesthetics and\nattributes database (AADB) which contains aesthetic scores and meaningful\nattributes assigned to each image by multiple human raters. Anonymized rater\nidentities are recorded across images allowing us to exploit intra-rater\nconsistency using a novel sampling strategy when computing the ranking loss of\ntraining image pairs. We show the proposed sampling strategy is very effective\nand robust in face of subjective judgement of image aesthetics by individuals\nwith different aesthetic tastes. Experiments demonstrate that our unified model\ncan generate aesthetic rankings that are more consistent with human ratings. To\nfurther validate our model, we show that by simply thresholding the estimated\naesthetic scores, we are able to achieve state-or-the-art classification\nperformance on the existing AVA dataset benchmark.","url_abs":"http://arxiv.org/abs/1606.01621v2","url_pdf":"http://arxiv.org/pdf/1606.01621v2.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":"photo-aesthetics-ranking-network-with","repo_url":"https://github.com/IsaacCorley/deep-aesthetics-pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"photo-aesthetics-ranking-network-with","repo_url":"https://github.com/aimerykong/deepImageAestheticsAnalysis","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"aesthetics-quality-assessment","task_name":"Aesthetics Quality Assessment"}],"methods":[],"datasets_introduced":[{"slug":"aadb","name":"AADB","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/aesthetics-quality-assessment-on-ava","task":"Aesthetics Quality Assessment","dataset":"AVA","model":"ADB-CNN","rank_in_archive_order":7,"of":9,"metrics":{"Accuracy":"77.3%"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1606.01621","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}