{"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/revisiting-single-image-depth-estimation","title":"Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps with Accurate Object Boundaries","arxiv_id":"1803.08673","date":"2018-03-23","proceeding":null,"authors":["Junjie Hu","Mete Ozay","Yan Zhang","Takayuki Okatani"],"abstract":"This paper considers the problem of single image depth estimation. The\nemployment of convolutional neural networks (CNNs) has recently brought about\nsignificant advancements in the research of this problem. However, most\nexisting methods suffer from loss of spatial resolution in the estimated depth\nmaps; a typical symptom is distorted and blurry reconstruction of object\nboundaries. In this paper, toward more accurate estimation with a focus on\ndepth maps with higher spatial resolution, we propose two improvements to\nexisting approaches. One is about the strategy of fusing features extracted at\ndifferent scales, for which we propose an improved network architecture\nconsisting of four modules: an encoder, decoder, multi-scale feature fusion\nmodule, and refinement module. The other is about loss functions for measuring\ninference errors used in training. We show that three loss terms, which measure\nerrors in depth, gradients and surface normals, respectively, contribute to\nimprovement of accuracy in an complementary fashion. Experimental results show\nthat these two improvements enable to attain higher accuracy than the current\nstate-of-the-arts, which is given by finer resolution reconstruction, for\nexample, with small objects and object boundaries.","url_abs":"http://arxiv.org/abs/1803.08673v2","url_pdf":"http://arxiv.org/pdf/1803.08673v2.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":"revisiting-single-image-depth-estimation","repo_url":"https://github.com/JunjH/Revisiting_Single_Depth_Estimation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"revisiting-single-image-depth-estimation","repo_url":"https://github.com/Seojiyoung/Depth-Map-Estimation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"revisiting-single-image-depth-estimation","repo_url":"https://github.com/Xt-Chen/SARPN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"revisiting-single-image-depth-estimation","repo_url":"https://github.com/karasawatakumi/REVISITING_MonoDepth","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"depth-estimation","task_name":"Depth Estimation"},{"task_slug":"monocular-depth-estimation","task_name":"Monocular Depth Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/monocular-depth-estimation-on-nyu-depth-v2","task":"Monocular Depth Estimation","dataset":"NYU-Depth V2","model":"SENet-154","rank_in_archive_order":74,"of":85,"metrics":{"RMSE":"0.530"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1803.08673","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}