Panasonic e Hlahisa Mekhoa e 'meli e tsoetseng pele ea AI,
E amohetsoe ho CVPR2021,
Khonferense e Eteletseng Pele ea Lefatše ea AI Technology
[1] Home Action Genome: Contrast Compositional Action Understanding
Re thabela ho phatlalatsa hore re thehile "dataset" e ncha ea "Home Action Genome" e bokellang mesebetsi ea letsatsi le letsatsi ea batho malapeng a bona ka mefuta e mengata ea li-sensor, ho kenyeletsoa lik'hamera, li-microphone le li-sensors tsa mocheso. Re thehile le ho lokolla pokello ea lintlha tse kholo ka ho fetisisa lefatšeng tsa mekhoa e mengata bakeng sa libaka tsa bolulo, athe boholo ba li-database tsa libaka tsa bolulo li nyenyane ka tekanyo. Ka ho sebelisa dataset ena, bafuputsi ba AI ba ka e sebelisa e le lintlha tsa koetliso bakeng sa ho ithuta mochine le lipatlisiso tsa AI ho tšehetsa batho ba sebakeng sa bolulo.
Ho phaella ho tse ka holimo, re thehile theknoloji ea ho ithuta ea tšebelisano-'moho bakeng sa kananelo ea mesebetsi ea maemo a phahameng ka ho fetisisa le maikutlo a mangata. Ka ho sebelisa thekenoloji ena, re ka ithuta likarolo tse lumellanang pakeng tsa maikutlo a fapaneng, li-sensor, boitšoaro ba maemo a phahameng, le li-label tse qaqileng tsa boitšoaro, 'me kahoo ra ntlafatsa ts'ebetso ea ho lemoha mesebetsi e rarahaneng libakeng tsa bolulo.
Theknoloji ena ke phello ea lipatlisiso tse entsoeng ka tšebelisano pakeng tsa Digital AI Technology Center, Technology Division, le Stanford Vision le Lab ea ho Ithuta Univesithing ea Stanford.
Setšoantšo sa 1: Kutloisiso ea Ts'ebetso ea Cooperative Compositional Action (CCAU) Ho koetlisa mekhoa eohle hammoho ho re fa monyetla oa ho bona ts'ebetso e ntlafetseng.
Re sebelisa lithupelo re sebelisa lileibole tsa boemo ba video le liathomo ho lumella livideo le liketso tsa athomo ho rua molemo ho tsoa ho litšebelisano lipakeng tsa tse peli.
[2] AutoDO: Robust AutoAugment bakeng sa Data e Biased le Label Noise ka Scalable Probabilistic Implicit Differentiation
Re boetse re thabetse ho phatlalatsa hore re thehile theknoloji e ncha ea ho ithuta ka mochini e iketsetsang ntlafatso e nepahetseng ea data ho latela kabo ea data ea koetliso. Theknoloji ena e ka sebelisoa maemong a sebele a lefats'e, moo data e fumanehang e nyenyane haholo. Ho na le linyeoe tse ngata libakeng tsa rona tse kholo tsa khoebo, moo ho leng thata ho sebelisa theknoloji ea AI ka lebaka la mefokolo ea lintlha tse fumanehang. Ka ho sebelisa theknoloji ena, ts'ebetso ea tokiso ea li-parameter tsa augmentation ea data e ka felisoa, 'me li-parameter li ka fetoloa ka bohona. Ka hona, ho ka lebelloa hore mefuta e mengata ea tšebeliso ea theknoloji ea AI e ka phatlalatsoa ka ho pharaletseng. Nakong e tlang, ka ho tsoela pele ho potlakisa lipatlisiso le nts'etsopele ea theknoloji ena, re tla sebeletsa ho lemoha theknoloji ea AI e ka sebelisoang libakeng tsa sebele tsa lefatše tse kang lisebelisoa le mekhoa e tloaelehileng. Theknoloji ena ke phello ea lipatlisiso tse entsoeng ke Digital AI Technology Center, Technology Division, AI Laboratory ea Panasonic R & D Company of America.
Setšoantšo sa 2: AutoDO e rarolla bothata ba ho eketsa lintlha (Leano le arolelanoang la DA bothata) .kabo ea lintlha tse eketsehileng tsa terene (boputsoa bo boputsoa) bo ka 'na ba se ke ba lumellana le data ea teko (bofubelu bo tiileng) sebakeng se patehileng:
"2" ha e eketsehe, ha "5" e eketsehile ho feta tekano. Ka lebaka leo, mekhoa ea pele e ke ke ea lumellana le kabo ea liteko mme qeto ea mohlophisi ea ithutoang f(θ) ha ea nepahala.
Lintlha tsa theknoloji ena li tla hlahisoa ho CVPR2021 (e tla tšoaroa ho tloha ka June 19th, 2017).
Molaetsa o kaholimo o tsoa ho webosaete ea semmuso ea Panasonic!
Nako ea poso: Jun-03-2021
