eprintid: 10060971 rev_number: 21 eprint_status: archive userid: 608 dir: disk0/10/06/09/71 datestamp: 2018-11-12 15:18:11 lastmod: 2021-11-23 23:29:58 status_changed: 2018-11-12 15:18:11 type: proceedings_section metadata_visibility: show creators_name: Neverova, N creators_name: Kokkinos, I title: Mass Displacement Networks ispublished: pub divisions: UCL divisions: B04 divisions: C05 divisions: F48 note: This version is the author accepted manuscript. For information on re-use, please refer to the publisher’s terms and conditions. abstract: Despite the large improvements in performance attained by deep learning in computer vision, one can often further improve results with some additional post-processing that exploits the geometric nature of the underlying task. This commonly involves displacing the posterior distribution of a CNN in a way that makes it more appropriate for the task at hand, e.g. better aligned with local image features, or more compact. In this work we integrate this geometric post-processing within a deep architecture, introducing a differentiable and probabilistically sound counterpart to the common geometric voting technique used for evidence accumulation in vision. We refer to the resulting neural models as Mass Displacement Networks (MDNs), and apply them to human pose estimation in two distinct setups: (a) landmark localization, where we collapse a distribution to a point, allowing for precise localization of body keypoints and (b) communication across body parts, where we transfer evidence from one part to the other, allowing for a globally consistent pose estimate. We evaluate on large-scale pose estimation benchmarks, such as MPII Human Pose and COCO datasets, and report systematic improvements. date: 2018-09 date_type: published publisher: BMVA Press official_url: http://bmvc2018.org/programmedetail.html oa_status: green full_text_type: other language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 1590197 lyricists_name: Kokkinos, Iason lyricists_id: IKOKK25 actors_name: Kokkinos, Iason actors_id: IKOKK25 actors_role: owner full_text_status: public series: British Machine Vision Conference publication: BMVC volume: 29 event_title: BMVC 2018, 29th British Machine Vision Conference, 3–6 September 2018, Newcastle upon Tyne, UK book_title: 29th British Machine Vision Conference (BMVC) 2018 editors_name: Shao, L editors_name: Shum, HPH editors_name: Hospedales, T citation: Neverova, N; Kokkinos, I; (2018) Mass Displacement Networks. In: Shao, L and Shum, HPH and Hospedales, T, (eds.) 29th British Machine Vision Conference (BMVC) 2018. BMVA Press Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10060971/1/0060.pdf