{"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/generic-3d-representation-via-pose-estimation","title":"Generic 3D Representation via Pose Estimation and Matching","arxiv_id":"1710.08247","date":"2017-10-23","proceeding":null,"authors":["Amir R. Zamir","Tilman Wekel","Pulkit Argrawal","Colin Weil","Jitendra Malik","Silvio Savarese"],"abstract":"Though a large body of computer vision research has investigated developing\ngeneric semantic representations, efforts towards developing a similar\nrepresentation for 3D has been limited. In this paper, we learn a generic 3D\nrepresentation through solving a set of foundational proxy 3D tasks:\nobject-centric camera pose estimation and wide baseline feature matching. Our\nmethod is based upon the premise that by providing supervision over a set of\ncarefully selected foundational tasks, generalization to novel tasks and\nabstraction capabilities can be achieved. We empirically show that the internal\nrepresentation of a multi-task ConvNet trained to solve the above core problems\ngeneralizes to novel 3D tasks (e.g., scene layout estimation, object pose\nestimation, surface normal estimation) without the need for fine-tuning and\nshows traits of abstraction abilities (e.g., cross-modality pose estimation).\nIn the context of the core supervised tasks, we demonstrate our representation\nachieves state-of-the-art wide baseline feature matching results without\nrequiring apriori rectification (unlike SIFT and the majority of learned\nfeatures). We also show 6DOF camera pose estimation given a pair local image\npatches. The accuracy of both supervised tasks come comparable to humans.\nFinally, we contribute a large-scale dataset composed of object-centric street\nview scenes along with point correspondences and camera pose information, and\nconclude with a discussion on the learned representation and open research\nquestions.","url_abs":"http://arxiv.org/abs/1710.08247v1","url_pdf":"http://arxiv.org/pdf/1710.08247v1.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":"generic-3d-representation-via-pose-estimation","repo_url":"https://github.com/amir32002/3D_Street_View","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"camera-pose-estimation","task_name":"Camera Pose Estimation"},{"task_slug":"object","task_name":"Object"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"},{"task_slug":"surface-normal-estimation","task_name":"Surface Normal Estimation"}],"methods":[],"datasets_introduced":[{"slug":"street-view-image-pose-and-3d-cities-dataset","name":"Street View Image, Pose, and 3D Cities Dataset","full_name":""}],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1710.08247","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}