{"id":"https://openalex.org/W2885567378","doi":"https://doi.org/10.5075/epfl-thesis-8677","title":"Mean-Field methods for Structured Deep-Learning in Computer Vision","display_name":"Mean-Field methods for Structured Deep-Learning in Computer Vision","publication_year":2018,"publication_date":"2018-01-01","ids":{"openalex":"https://openalex.org/W2885567378","doi":"https://doi.org/10.5075/epfl-thesis-8677","mag":"2885567378"},"language":"en","primary_location":{"id":"pmh:oai:infoscience.epfl.ch:255942","is_oa":true,"landing_page_url":"https://infoscience.epfl.ch/handle/20.500.14299/147197","pdf_url":"https://infoscience.epfl.ch/record/255942/files/EPFL_TH8677.pdf","source":{"id":"https://openalex.org/S4306400487","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"doctoral thesis"},"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://infoscience.epfl.ch/record/255942/files/EPFL_TH8677.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5030971556","display_name":"Pierre Baqu\u00e9","orcid":"https://orcid.org/0000-0002-0294-1408"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Baqu\u00e9, Pierre Bruno","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5030971556"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10688","display_name":"Image and Signal Denoising Methods","score":0.7361000180244446,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10688","display_name":"Image and Signal Denoising Methods","score":0.7361000180244446,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.6225359439849854},{"id":"https://openalex.org/keywords/field","display_name":"Field (mathematics)","score":0.5460247993469238},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5453973412513733},{"id":"https://openalex.org/keywords/computer-vision","display_name":"Computer vision","score":0.507225751876831},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.47221094369888306},{"id":"https://openalex.org/keywords/computer-graphics","display_name":"Computer graphics (images)","score":0.36717501282691956},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.18632152676582336}],"concepts":[{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6225359439849854},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.5460247993469238},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5453973412513733},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.507225751876831},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.47221094369888306},{"id":"https://openalex.org/C121684516","wikidata":"https://www.wikidata.org/wiki/Q7600677","display_name":"Computer graphics (images)","level":1,"score":0.36717501282691956},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.18632152676582336},{"id":"https://openalex.org/C202444582","wikidata":"https://www.wikidata.org/wiki/Q837863","display_name":"Pure mathematics","level":1,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:infoscience.epfl.ch:255942","is_oa":true,"landing_page_url":"https://infoscience.epfl.ch/handle/20.500.14299/147197","pdf_url":"https://infoscience.epfl.ch/record/255942/files/EPFL_TH8677.pdf","source":{"id":"https://openalex.org/S4306400487","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"doctoral thesis"},{"id":"doi:10.5075/epfl-thesis-8677","is_oa":true,"landing_page_url":"https://doi.org/10.5075/epfl-thesis-8677","pdf_url":null,"source":{"id":"https://openalex.org/S4306400488","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"thesis"},{"id":"mag:2885567378","is_oa":false,"landing_page_url":"https://infoscience.epfl.ch/record/255942","pdf_url":null,"source":null,"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":null}],"best_oa_location":{"id":"pmh:oai:infoscience.epfl.ch:255942","is_oa":true,"landing_page_url":"https://infoscience.epfl.ch/handle/20.500.14299/147197","pdf_url":"https://infoscience.epfl.ch/record/255942/files/EPFL_TH8677.pdf","source":{"id":"https://openalex.org/S4306400487","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"doctoral thesis"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W2885567378.pdf","grobid_xml":"https://content.openalex.org/works/W2885567378.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2586272128","https://openalex.org/W2963479620","https://openalex.org/W3099488366","https://openalex.org/W2885371173","https://openalex.org/W2963720057","https://openalex.org/W2759424660","https://openalex.org/W2944123037","https://openalex.org/W3213766945","https://openalex.org/W3101183984","https://openalex.org/W3196563995","https://openalex.org/W3116836465","https://openalex.org/W3035488168","https://openalex.org/W3143863175","https://openalex.org/W206687385","https://openalex.org/W605922786","https://openalex.org/W3039897977","https://openalex.org/W2163055326","https://openalex.org/W2980529652","https://openalex.org/W2517999587","https://openalex.org/W2005829989"],"abstract_inverted_index":{"In":[0,102],"recent":[1],"years,":[2],"Machine":[3],"Learning":[4],"based":[5,100,115,149,335,379],"Computer":[6],"Vision":[7],"techniques":[8],"made":[9],"impressive":[10],"progress.":[11],"These":[12],"algorithms":[13,278,293],"proved":[14],"particularly":[15],"efficient":[16,327],"for":[17,331,368],"image":[18,66],"classification":[19],"or":[20,36],"detection":[21,62,370,382],"of":[22,47,78,98,217,223,237,254,268,285,289,315,363,387],"isolated":[23],"objects.":[24],"From":[25],"a":[26,110,122,134,163,220,250,266,324,377],"probabilistic":[27],"perspective,":[28],"these":[29,92],"methods":[30,350],"can":[31,157,211,342],"predict":[32,53],"marginals,":[33],"over":[34,140],"single":[35],"multiple":[37,313],"variables,":[38],"independently,":[39],"with":[40],"high":[41],"accuracy.":[42],"However,":[43],"in":[44,63,82,131,154,189,242,357,371],"many":[45,155],"tasks":[46],"practical":[48,113],"interest,":[49],"we":[50,105,207,210,355],"need":[51],"to":[52,91,137,168,179,231,247,306,346],"jointly":[54],"several":[55],"correlated":[56],"variables.":[57],"Practical":[58],"applications":[59],"include":[60],"people":[61],"crowded":[64,372],"scenes,":[65],"segmentation,":[67],"surface":[68],"reconstruction,":[69],"3D":[70],"pose":[71],"estimation":[72],"and":[73,112,153,177,185,195,326,366],"others.":[74],"A":[75],"large":[76],"part":[77],"the":[79,96,170,213,229,232,238,255,261,286,290,296,316,318,337,360,388],"research":[80],"effort":[81],"today's":[83],"computer-vision":[84],"community":[85],"aims":[86],"at":[87],"finding":[88],"task-specific":[89],"solutions":[90],"problems,":[93],"while":[94],"leveraging":[95],"power":[97],"Deep-Learning":[99],"classifiers.":[101],"this":[103,274,392],"thesis,":[104],"present":[106,376],"our":[107],"journey":[108],"towards":[109],"generic":[111],"solution":[114],"on":[116,150,282,336,391],"mean-field":[117,146,203,218,269,291,378],"(MF)":[118],"inference.":[119],"Mean-field":[120],"is":[121,148,174,294],"Statistical":[123],"Physics-inspired":[124],"method":[125,240,330],"which":[126,341],"has":[127],"long":[128],"been":[129],"used":[130],"Computer-Vision":[132],"as":[133,352],"variational":[135,171],"approximation":[136,253],"posterior":[138,257,333],"distributions":[139],"complex":[141],"Conditional":[142],"Random":[143],"Fields.":[144],"Standard":[145],"optimization":[147,204],"coordinate":[151],"descent":[152],"situations":[156],"be":[158,343],"impractical.":[159],"We":[160,181,271,322,374],"therefore":[161],"propose":[162],"novel":[164,325],"proximal":[165],"gradient-based":[166,348],"approach":[167],"optimizing":[169],"objective.":[172],"It":[173],"naturally":[175,304],"parallelizable":[176],"easy":[178],"implement.":[180],"prove":[182],"its":[183],"convergence,":[184],"then":[186,375],"demonstrate":[187,272],"that,":[188],"practice,":[190],"it":[191],"yields":[192],"faster":[193],"convergence":[194],"often":[196],"finds":[197],"better":[198],"optima":[199],"than":[200],"more":[201,251,358],"traditional":[202],"techniques.":[205],"Then,":[206],"show":[208],"that":[209,226,273,279,295,384],"replace":[212],"fully":[214,300],"factorized":[215],"distribution":[216,334],"by":[219,260,310],"weighted":[221],"mixture":[222,267],"such":[224,351],"distributions,":[225],"similarly":[227],"minimizes":[228],"KL-Divergence":[230],"true":[233,256],"posterior.":[234],"Our":[235],"extension":[236],"clamping":[239],"proposed":[241],"previous":[243],"works":[244],"allows":[245,305],"us":[246],"both":[248],"produce":[249],"descriptive":[252],"and,":[258],"inspired":[259],"diverse":[262],"MAP":[263],"paradigms,":[264],"fit":[265],"approximations.":[270],"positively":[275],"impacts":[276],"real-world":[277],"initially":[280],"relied":[281],"mean-fields.":[283],"One":[284],"important":[287],"properties":[288],"inference":[292],"closed-form":[297],"updates":[298],"are":[299],"differentiable":[301],"operations.":[302],"This":[303],"do":[307],"parameter":[308],"learning":[309,329,349,365],"simply":[311],"unrolling":[312],"iterations":[314],"updates,":[317],"so-called":[319],"back-mean-field":[320],"algorithm.":[321],"derive":[323],"structured":[328,364,380],"multi-modal":[332],"Multi-Modal":[338],"Mean-Field":[339],"approximation,":[340],"seamlessly":[344],"combined":[345],"modern":[347],"CNNs.":[353],"Finally,":[354],"explore":[356],"details":[359],"specific":[361],"problem":[362],"prediction":[367],"multiple-people":[369],"scenes.":[373],"deep-learning":[381],"algorithm":[383],"provides":[385],"state":[386],"art":[389],"results":[390],"dataset.":[393]},"counts_by_year":[],"updated_date":"2025-11-06T06:51:31.235846","created_date":"2025-10-10T00:00:00"}