{"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/t2net-synthetic-to-realistic-translation-for","title":"T2Net: Synthetic-to-Realistic Translation for Solving Single-Image Depth Estimation Tasks","arxiv_id":"1808.01454","date":"2018-08-04","proceeding":"ECCV 2018 9","authors":["Chuanxia Zheng","Tat-Jen Cham","Jianfei Cai"],"abstract":"Current methods for single-image depth estimation use training datasets with\nreal image-depth pairs or stereo pairs, which are not easy to acquire. We\npropose a framework, trained on synthetic image-depth pairs and unpaired real\nimages, that comprises an image translation network for enhancing realism of\ninput images, followed by a depth prediction network. A key idea is having the\nfirst network act as a wide-spectrum input translator, taking in either\nsynthetic or real images, and ideally producing minimally modified realistic\nimages. This is done via a reconstruction loss when the training input is real,\nand GAN loss when synthetic, removing the need for heuristic\nself-regularization. The second network is trained on a task loss for synthetic\nimage-depth pairs, with extra GAN loss to unify real and synthetic feature\ndistributions. Importantly, the framework can be trained end-to-end, leading to\ngood results, even surpassing early deep-learning methods that use real paired\ndata.","url_abs":"http://arxiv.org/abs/1808.01454v1","url_pdf":"http://arxiv.org/pdf/1808.01454v1.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":"t2net-synthetic-to-realistic-translation-for","repo_url":"https://github.com/lyndonzheng/Synthetic2Realistic","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"depth-estimation","task_name":"Depth Estimation"},{"task_slug":"depth-prediction","task_name":"Depth Prediction"},{"task_slug":"translation","task_name":"Translation"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/depth-estimation-on-dcm","task":"Depth Estimation","dataset":"DCM","model":"T2Net","rank_in_archive_order":3,"of":3,"metrics":{"Abs Rel":"0.351","RMSE":"1.117","RMSE log":"0.415","Sq Rel":"0.416"},"uses_additional_data":false},{"leaderboard":"/sota/depth-estimation-on-ebdtheque","task":"Depth Estimation","dataset":"eBDtheque","model":"T2Net","rank_in_archive_order":3,"of":3,"metrics":{"Abs Rel":"0.491","RMSE":"1.459","RMSE log":"0.777","Sq Rel":"0.555"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-domain-adaptation-on-virtual-2","task":"Unsupervised Domain Adaptation","dataset":"virtual KITTI to KITTI (MDE)","model":"T2Net","rank_in_archive_order":3,"of":4,"metrics":{"RMSE ":"4.674"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1808.01454","atlas_url":"https://app.syntology.ai/?focus=1808.01454","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}