{"id":350834,"date":"2017-01-11T12:02:13","date_gmt":"2017-01-11T20:02:13","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-research-item&p=350834"},"modified":"2018-10-16T19:55:52","modified_gmt":"2018-10-17T02:55:52","slug":"towards-reversal-invariant-image-representation","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/towards-reversal-invariant-image-representation\/","title":{"rendered":"Towards Reversal-Invariant Image Representation"},"content":{"rendered":"
State-of-the-art<\/span> image<\/span> classi<\/span>\ufb01<\/span>cation<\/span> approaches<\/span><\/span><\/div>\n
are<\/span> mainly<\/span> based<\/span> on<\/span> rob<\/span>ust<\/span> image<\/span> representation,<\/span> such<\/span> as<\/span><\/div>\n
the<\/span> bag-of-features<\/span> (BoF)<\/span> model<\/span> or<\/span> the<\/span> con<\/span>v<\/span>olutional<\/span> neural<\/span><\/div>\n
network<\/span> (CNN)<\/span> architecture.<\/span> In<\/span> real<\/span> applications,<\/span> the<\/span> ori-<\/span><\/div>\n
entation<\/span> (left\/right)<\/span> of<\/span> an<\/span> image<\/span> or<\/span> an<\/span> object<\/span> might<\/span> v<\/span>ary<\/span><\/span><\/div>\n
from<\/span> sample<\/span> to<\/span> sample,<\/span> whereas<\/span> some<\/span> handcrafted<\/span> descrip-<\/span><\/div>\n
tors<\/span> (e.g.,<\/span> SIFT)<\/span> and<\/span> network<\/span> operations<\/span> (e.g.,<\/span> con<\/span>v<\/span>olution)<\/span><\/div>\n
are<\/span> not<\/span> re<\/span>versal-in<\/span>v<\/span>ariant,<\/span> leading<\/span> to<\/span> the<\/span> unsatis<\/span>\ufb01<\/span>ed<\/span> stabil-<\/span><\/div>\n
ity<\/span> of<\/span> image<\/span> features<\/span> e<\/span>xtracted<\/span> from<\/span> these<\/span> models.<\/span> T<\/span>o<\/span> deal<\/span><\/div>\n
with,<\/span> a<\/span> popular<\/span> solution<\/span> is<\/span> to<\/span> augment<\/span> the<\/span> dataset<\/span> by<\/span> adding<\/span><\/div>\n
a<\/span> left-right<\/span> re<\/span>v<\/span>ersed<\/span> copy<\/span> for<\/span> each<\/span> image.<\/span> This<\/span> strate<\/span>gy<\/span><\/div>\n
impro<\/span>v<\/span>es<\/span> the<\/span> recognition<\/span> accurac<\/span>y<\/span> to<\/span> some<\/span> extent,<\/span> b<\/span>ut<\/span> also\u00a0<\/span><\/span><\/p>\n
brings<\/span> the<\/span> price<\/span> of<\/span> almost<\/span> doubled<\/span> time<\/span> and<\/span> memory<\/span> con-<\/span><\/div>\n
sumptions<\/span> on<\/span> both<\/span> the<\/span> training<\/span> and<\/span> testing<\/span> stages.<\/span> In<\/span> this<\/span><\/div>\n
paper<\/span>,<\/span> we<\/span> present<\/span> an<\/span> alternati<\/span>v<\/span>e<\/span> solution<\/span> based<\/span> on<\/span> design-<\/span><\/div>\n
ing<\/span> re<\/span>v<\/span>ersal-in<\/span>v<\/span>ariant<\/span> representation<\/span> of<\/span> local<\/span> patterns,<\/span> so<\/span> that<\/span><\/div>\n
we<\/span> can<\/span> obtain<\/span> the<\/span> identical<\/span> representation<\/span> for<\/span> an<\/span> image<\/span> and<\/span><\/div>\n
its<\/span> left-right<\/span> re<\/span>versed<\/span> copy<\/span>.<\/span> F<\/span>or<\/span> the<\/span> BoF<\/span> model,<\/span> we<\/span> design<\/span><\/div>\n
a<\/span> re<\/span>v<\/span>ersal-in<\/span>v<\/span>ariant<\/span> v<\/span>ersion<\/span> of<\/span> SIFT<\/span> descriptor<\/span> named<\/span> Max-<\/span><\/div>\n
SIFT<\/span>,<\/span> a<\/span> generalized<\/span> <\/span>RIDE<\/span> algorithm<\/span> which<\/span> can<\/span> be<\/span> applied<\/span> to<\/span><\/span><\/div>\n
a<\/span> lar<\/span>ge<\/span> family<\/span> of<\/span> local<\/span> descriptors.<\/span> For<\/span> the<\/span> CNN<\/span> architecture,<\/span><\/div>\n
we<\/span> present<\/span> a<\/span> simple<\/span> idea<\/span> of<\/span> generating<\/span> re<\/span>v<\/span>ersal-in<\/span>v<\/span>ariant<\/span> deep<\/span><\/div>\n
features<\/span> (<\/span>RI-Deep<\/span>),<\/span> and,<\/span> inspired<\/span> by<\/span> which,<\/span> design<\/span> re<\/span>versal-<\/span><\/div>\n
in<\/span>v<\/span>ariant<\/span> con<\/span>v<\/span>olution<\/span> (<\/span>RI-Con<\/span>v<\/span>)<\/span> layers<\/span> to<\/span> increase<\/span> the<\/span> CNN<\/span><\/span><\/span><\/div>\n
capacity<\/span> without<\/span> increasing<\/span> the<\/span> model<\/span> complexity<\/span>.<\/span> Experi-<\/span><\/div>\n
ments<\/span> re<\/span>veal<\/span> consistent<\/span> accuracy<\/span> gain<\/span> on<\/span> v<\/span>arious<\/span> image<\/span> clas-<\/span><\/div>\n
si<\/span>\ufb01<\/span>cation<\/span> tasks,<\/span> including<\/span> scene<\/span> understanding,<\/span> \ufb01<\/span>ne-grained<\/span><\/div>\n
object<\/span> recognition,<\/span> and<\/span> lar<\/span>ge-scale<\/span> visual<\/span> recognition.<\/span><\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

State-of-the-art image classi\ufb01cation approaches are mainly based on robust image representation, such as the bag-of-features (BoF) model or the convolutional neural network (CNN) architecture. In real applications, the ori- entation (left\/right) of an image or an object might vary from sample to sample, whereas some handcrafted descrip- tors (e.g., SIFT) and network operations (e.g., convolution) […]<\/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":"","msr-author-ordering":null,"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"","msr_doi":"","msr_arxiv_id":"","msr_s2_paper_id":"","msr_mag_id":"","msr_pubmed_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_original_fields_of_study":"","msr_release_tracker_id":"","msr_s2_match_type":"","msr_citation_count_updated":"","msr_published_date":"2016-11-01","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"http:\/\/www.readcube.com\/articles\/10.1007\/s11263-016-0970-x?author_access_token=x9u_Aw0xjT6J4io7TeIJAfe4RwlQNchNByi7wbcMAY5iFYjMXQvCQfhC7tl6PSjZkSYiAsOFV18-KBKnRgmA-fRyOonZ8liD8IAjGE5uyOBM3XFHHbDIqr1_TWg2S_bBaBvQfjuyJGj86FxkbuiLeA%3D%3D","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_s2_pdf_url":"","msr_year":0,"msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":0,"msr_s2_match_confidence":0,"msr_microsoftintellectualproperty":true,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13562],"msr-publication-type":[193715],"msr-publisher":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-350834","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-computer-vision","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2016-11-01","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"http:\/\/www.readcube.com\/articles\/10.1007\/s11263-016-0970-x?author_access_token=x9u_Aw0xjT6J4io7TeIJAfe4RwlQNchNByi7wbcMAY5iFYjMXQvCQfhC7tl6PSjZkSYiAsOFV18-KBKnRgmA-fRyOonZ8liD8IAjGE5uyOBM3XFHHbDIqr1_TWg2S_bBaBvQfjuyJGj86FxkbuiLeA%3D%3D","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"http:\/\/www.readcube.com\/articles\/10.1007\/s11263-016-0970-x?author_access_token=x9u_Aw0xjT6J4io7TeIJAfe4RwlQNchNByi7wbcMAY5iFYjMXQvCQfhC7tl6PSjZkSYiAsOFV18-KBKnRgmA-fRyOonZ8liD8IAjGE5uyOBM3XFHHbDIqr1_TWg2S_bBaBvQfjuyJGj86FxkbuiLeA%3D%3D","viewUrl":false,"id":false,"label_id":0}],"msr_related_uploader":"","msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"","msr_influential_citations":0,"msr_reference_count":0,"msr_arxiv_id":"","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[{"id":0,"url":"http:\/\/www.readcube.com\/articles\/10.1007\/s11263-016-0970-x?author_access_token=x9u_Aw0xjT6J4io7TeIJAfe4RwlQNchNByi7wbcMAY5iFYjMXQvCQfhC7tl6PSjZkSYiAsOFV18-KBKnRgmA-fRyOonZ8liD8IAjGE5uyOBM3XFHHbDIqr1_TWg2S_bBaBvQfjuyJGj86FxkbuiLeA%3D%3D"}],"msr-author-ordering":[{"type":"user_nicename","value":"jingdw","user_id":32299,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=jingdw"},{"type":"text","value":"Lingxi Xie","user_id":0,"rest_url":false},{"type":"text","value":"Weiyao Lin","user_id":0,"rest_url":false},{"type":"text","value":"Bo Zhang","user_id":0,"rest_url":false},{"type":"text","value":"Qi Tian","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"article","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/350834","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":1,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/350834\/revisions"}],"predecessor-version":[{"id":512735,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/350834\/revisions\/512735"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=350834"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=350834"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=350834"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=350834"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=350834"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=350834"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=350834"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=350834"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=350834"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=350834"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=350834"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=350834"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=350834"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}