{"id":285590,"date":"2016-08-30T20:04:07","date_gmt":"2016-08-31T03:04:07","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&p=285590"},"modified":"2017-05-31T11:17:04","modified_gmt":"2017-05-31T18:17:04","slug":"3d-object-reconstruction-recognition","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/3d-object-reconstruction-recognition\/","title":{"rendered":"3D Object Reconstruction and Recognition"},"content":{"rendered":"
We study the problem of 3D object reconstruction and recognition. For reconstruction, we aim at developing algorithms and systems to lower down the barrier of 3D reconstruction for common users. In this way, we can collect a world-class 3D object repository via leveraging crowdsourcing. For recognition, we aim at dealing with a large-scale task (e.g. identifying thousands of objects), and providing real-time performance.<\/p>\n<\/div>\n
coming soon …<\/p>\n
<\/p>\n
<\/p>\n
3D model-based object recognition has been a noticeable research trend in recent years. Common methods identify 2D-to-3D correspondences and make recognition decisions by RANSAC-based pose estimation, whose efficiency usually suffers from inaccurate correspondences caused by the increasing number of target objects for recognition. To overcome this scalability bottleneck, in this paper we propose an efficient 2D-to-3D correspondence filtering approach which combines a light-weight neighborhood-based step with a finer-grained pairwise step to remove spurious
\ncorrespondences by 2D\/3D geometric cues. On a dataset of 300 3D objects, the proposed solution achieves \uff5e10 times speed improvement over the baseline, with a comparable recognition accuracy. A non-GPU implementation achieves a speed of \uff5e3fps for 1280\u00d7720 query images.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"
We study the problem of 3D object reconstruction and recognition. For reconstruction, we aim at developing algorithms and systems to lower down the barrier of 3D reconstruction for common users. In this way, we can collect a world-class 3D object repository via leveraging crowdsourcing. For recognition, we aim at dealing with a large-scale task (e.g. […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"research-area":[13551],"msr-locale":[268875],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-285590","msr-project","type-msr-project","status-publish","hentry","msr-research-area-graphics-and-multimedia","msr-locale-en_us","msr-archive-status-active"],"msr_project_start":"2015-05-01","related-publications":[],"related-downloads":[],"related-videos":[],"related-groups":[144916],"related-events":[],"related-opportunities":[],"related-posts":[],"related-articles":[],"tab-content":[{"id":0,"name":"Dataset","content":"