{"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/detect-replace-refine-deep-structured","title":"Detect, Replace, Refine: Deep Structured Prediction For Pixel Wise Labeling","arxiv_id":"1612.04770","date":"2016-12-14","proceeding":"CVPR 2017 7","authors":["Spyros Gidaris","Nikos Komodakis"],"abstract":"Pixel wise image labeling is an interesting and challenging problem with\ngreat significance in the computer vision community. In order for a dense\nlabeling algorithm to be able to achieve accurate and precise results, it has\nto consider the dependencies that exist in the joint space of both the input\nand the output variables. An implicit approach for modeling those dependencies\nis by training a deep neural network that, given as input an initial estimate\nof the output labels and the input image, it will be able to predict a new\nrefined estimate for the labels. In this context, our work is concerned with\nwhat is the optimal architecture for performing the label improvement task. We\nargue that the prior approaches of either directly predicting new label\nestimates or predicting residual corrections w.r.t. the initial labels with\nfeed-forward deep network architectures are sub-optimal. Instead, we propose a\ngeneric architecture that decomposes the label improvement task to three steps:\n1) detecting the initial label estimates that are incorrect, 2) replacing the\nincorrect labels with new ones, and finally 3) refining the renewed labels by\npredicting residual corrections w.r.t. them. Furthermore, we explore and\ncompare various other alternative architectures that consist of the\naforementioned Detection, Replace, and Refine components. We extensively\nevaluate the examined architectures in the challenging task of dense disparity\nestimation (stereo matching) and we report both quantitative and qualitative\nresults on three different datasets. Finally, our dense disparity estimation\nnetwork that implements the proposed generic architecture, achieves\nstate-of-the-art results in the KITTI 2015 test surpassing prior approaches by\na significant margin.","url_abs":"http://arxiv.org/abs/1612.04770v1","url_pdf":"http://arxiv.org/pdf/1612.04770v1.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":"detect-replace-refine-deep-structured","repo_url":"https://github.com/gidariss/DRR_struct_pred","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"disparity-estimation","task_name":"Disparity Estimation"},{"task_slug":"stereo-matching-1","task_name":"Stereo Matching"},{"task_slug":"stereo-matching","task_name":"Stereo Matching Hand"},{"task_slug":"structured-prediction","task_name":"Structured Prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1612.04770","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}