{"id":193,"date":"2023-02-22T14:49:32","date_gmt":"2023-02-22T13:49:32","guid":{"rendered":"https:\/\/project.inria.fr\/dare\/?page_id=193"},"modified":"2023-06-22T21:08:14","modified_gmt":"2023-06-22T19:08:14","slug":"title-semantic-alignment-for-multi-item-compression","status":"publish","type":"page","link":"https:\/\/project.inria.fr\/dare\/publications\/title-semantic-alignment-for-multi-item-compression\/","title":{"rendered":"Title: Semantic alignment for multi-item compression"},"content":{"rendered":"\n<p><strong>Authors: <\/strong>Tom Bachard,\u00a0 Anju Jose Tom, Thomas Maugey<strong><br \/><\/strong><\/p>\n<p><strong>Abstract: <\/strong><\/p>\n<p style=\"text-align: left;\"><em><span dir=\"ltr\" role=\"presentation\">Coding algorithms usually compress independently the im<\/span><span dir=\"ltr\" role=\"presentation\">ages of a collection, in particular when the correlation be<\/span><span dir=\"ltr\" role=\"presentation\">tween them only resides at the semantic level,<\/span> <span dir=\"ltr\" role=\"presentation\">i.e.,<\/span> <span dir=\"ltr\" role=\"presentation\">informa<\/span><span dir=\"ltr\" role=\"presentation\">tion related to the high-level image content. In this work, we<\/span> <span dir=\"ltr\" role=\"presentation\">propose a coding solution able to exploit this semantic redun<\/span><span dir=\"ltr\" role=\"presentation\">dancy to decrease the storage cost of data collections. First<\/span> <span dir=\"ltr\" role=\"presentation\">we introduce the multi-item compression framework.<\/span> <span dir=\"ltr\" role=\"presentation\">Then<\/span> <span dir=\"ltr\" role=\"presentation\">we derive a loss term to shape the latent space of a varia<\/span><span dir=\"ltr\" role=\"presentation\">tional auto-encoder so that the latent vectors of semantically<\/span> <span dir=\"ltr\" role=\"presentation\">identical images can be aligned. Finally, we experimentally<\/span> <span dir=\"ltr\" role=\"presentation\">demonstrate that this alignment leads to a more compact rep<\/span><span dir=\"ltr\" role=\"presentation\">resentation of the data collection.<\/span><\/em><\/p>\n<p><strong>Contributions: <\/strong><\/p>\n<p><strong>WIP<br \/><\/strong><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Authors: Tom Bachard,\u00a0 Anju Jose Tom, Thomas Maugey Abstract: Coding algorithms usually compress independently the images of a collection, in particular when the correlation between them only resides at the semantic level, i.e., information related to the high-level image content. In this work, we propose a coding solution able to\u2026<\/p>\n<p> <a class=\"continue-reading-link\" href=\"https:\/\/project.inria.fr\/dare\/publications\/title-semantic-alignment-for-multi-item-compression\/\"><span>Continue reading<\/span><i class=\"crycon-right-dir\"><\/i><\/a> <\/p>\n","protected":false},"author":2222,"featured_media":0,"parent":119,"menu_order":2,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-193","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/pages\/193","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/users\/2222"}],"replies":[{"embeddable":true,"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/comments?post=193"}],"version-history":[{"count":3,"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/pages\/193\/revisions"}],"predecessor-version":[{"id":216,"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/pages\/193\/revisions\/216"}],"up":[{"embeddable":true,"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/pages\/119"}],"wp:attachment":[{"href":"https:\/\/project.inria.fr\/dare\/wp-json\/wp\/v2\/media?parent=193"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}