Kutheni ukuqondwa komfanekiso kubalulekile?

Malunga neepesenti ze-80 zomxholo kwi-intanethi zibonwa. Usenokuqala ukukhangela ukuba kutheni ukuthegiswa kwemifanekiso kunokubamba indawo yayo njengenkosi yetafile yomxholo. Nokuba ngaba ngabantu okanye iinkampani, ukuqondwa komfanekiso we-AI kwenze ukuba kube lula ukuchonga izinto ezibonwayo kwi-Intanethi kunye nengxabano encinci. Kukho malunga ne-657 yeebhiliyoni zeefoto ezithunyelwa rhoqo ngonyaka ngedijithali, uninzi luvela kumajelo asekuhlaleni. I-chunk elungileyo yaloo mifanekiso ngabantu abakhuthaza iimveliso, nokuba benza njalo bengazi. Umxholo owenziwe ngumsebenzisi (i-UGC) kwifomu yayo ecocekileyo yinto ebalaseleyo yokuvumela iibrendi njengoko ibonelela ngeyona ndlela ilungileyo yokunyusa.
Kukho izixhobo zokuthengisa zokulumkisa iinkampani xa kukho umthengi okhankanyiweyo kumajelo eendaba ezentlalo, kodwa kuthekani xa ukunyuswa kweemveliso kwenzeka ngaphandle kokuba nabani na abhale igama lakhe kwiposti yentlalo? Kulapho ukuqondwa komfanekiso we-AI kungqina ixabiso layo. Ukuba iteknoloji yondliwe iiseti zedatha ezichanekileyo, i-AI inokuchonga umfanekiso ngaphandle kokukhankanywa kwethegi ethile. Iziphumo zixabiseke kakhulu ukuba iibrendi zilandele kwaye zilandele ukukhankanywa kwazo kwezentlalo.

Ingaba ukuqondwa komfanekiso kusebenza njani?

Njengoko sisazi ukuba i-AI inokukhangela amaqonga eendaba ezentlalo ekhangela iifoto kwaye ithelekise neeseti zedatha ebanzi. Emva koko ithatha isigqibo ngomfanekiso ofanelekileyo ohambelana nesantya esikhawuleza kakhulu kunokuba abantu banako. Iimveliso zisebenzisa ukuqondwa komfanekiso ukufumana umxholo ofanayo nowabo kumajelo asekuhlaleni. Oko kuthetha ukuchonga uphawu lophawu okanye ukuqaphela ukubekwa kwemveliso phakathi kwabasebenzisi bemidiya yoluntu. Ukucela abantu ukuba bajonge ulwazi oluninzi kulula ngokulula. I-AI ayinakukhathazeka ngempazamo yomntu, kwaye ibuyisela iziphumo ezichanekileyo kumanqanaba angenakulinganiswa. Ukuqatshelwa komfanekiso we-AI kubeka esweni ukuba abantu bathini malunga ne-brand ngaphandle kwesidingo sesicatshulwa. Iibrendi ezikwaziyo ukulandelela ukukhankanywa kwazo kwezentlalo ngaphandle kwabasebenzisi abafuna ukuchwetheza igama lenkampani baya kuzifumana bekwindawo enenzuzo. Amandla okungena kukhuselo lwawo lwe-intanethi kuphela ngokusebenzisa izichongi ezaziwayo ze-AI zikhulu kwaye zibonelela ngokhuselo olungenakuthelekiswa nanto.

Nantsi eminye imisebenzi eqhelekileyo yokuqaphela umfanekiso:-

Ekuqaleni kufuneka sigqibe ukuba ingaba idatha yomfanekiso iqulethe into ethile, uphawu, okanye umsebenzi othile. Lo msebenzi unokusonjululwa ngamandla kwaye ngaphandle komzamo womntu, kodwa awukasombululwa ngokwanelisayo kumbono wekhompyuter kwimeko eqhelekileyo: izinto ezingafanelekanga kwiimeko ezingafanelekanga. Iindlela ezikhoyo zokujongana nale ngxaki zinokusombululwa ngcono kuphela kwizinto ezithile, njengezinto ezilula zejometri (umzekelo, i-polyhedra), ubuso bomntu, abalinganiswa abaprintiweyo okanye ababhalwe ngesandla, okanye izithuthi, kwaye kwiimeko ezithile, ezichazwe ngokuqhelekileyo ngokwemiqathango. yokukhanya okuchazwe kakuhle, imvelaphi, kunye nokuma kwento enxulumene nekhamera. Iindidi ezahlukeneyo zengxaki yokuqonda zichazwe kuncwadi:

• Ukuqatshelwa kwento

Into enye okanye ezininzi ezichazwe kwangaphambili okanye ezifundiweyo okanye iiklasi zento zinokuqatshelwa, ngokuqhelekileyo kunye neendawo zazo ze-2D emfanekisweni okanye i-3D yokumisa kwindawo yesiganeko.

• Ukuchongwa
Imeko yomntu ngamnye yento yamkelwa. Imizekelo kukuchongwa kobuso bomntu othile okanye umnwe weminwe, okanye ukuchongwa kwesithuthi esithile.

• Ukufunyanwa
Idatha yomfanekiso iskenwa kwimeko ethile. Imizekelo kukufunyanwa kweeseli ezingaqhelekanga okanye izicubu kwimifanekiso yezonyango okanye ukufunyanwa kwesithuthi kwindlela yokurhafisa yendlela ezenzekelayo. Ukufunyanwa okusekwe kuqikelelo olulula nolukhawulezayo lwezibalo ngamanye amaxesha kusetyenziswa ukufumana imimandla emincinci yedatha yemifanekiso enomdla enokuthi ihlalutywe ngakumbi ngobuchule obufuna kakhulu ukuvelisa utoliko oluchanekileyo.

Kukho imisebenzi eliqela ekhethekileyo esekwe kuqatshelwa, efana nale:

• Ukubuyiswa kwemifanekiso esekelwe kumxholo
Apha ukufumana yonke imifanekiso kwiseti enkulu yemifanekiso enomxholo othile. Umxholo ungacaciswa ngeendlela ezahlukeneyo, umzekelo ngokufana ngokuzalana nomfanekiso ekujoliswe kuwo (ndinike yonke imifanekiso efana nomfanekiso X), okanye ngokwemigaqo yokukhangela ekumgangatho ophezulu onikiweyo njengegalelo lombhalo (ndinike yonke imifanekiso equlathe izindlu ezininzi, zithathwa ebusika, kwaye azinamoto kuzo).

• Misa uqikelelo
kufuneka siqikelele indawo okanye ukuqhelaniswa kwento ethile ngokunxulumene nekhamera. Umzekelo wokusetyenziswa kobu buchule kuya kunceda i-robot ukukhupha izinto kwibhanti yokuhambisa kwimeko yomgca wendibano.

• Ukuqaphela abalinganiswa
I-OCR elichonga abalinganiswa kwimifanekiso yesicatshulwa esishicilelweyo okanye esibhaliweyo ngesandla, ngokuqhelekileyo ngembono yokufaka ikhowudi yokubhaliweyo kwifomathi ngakumbi kwaye yenza ukuhlela okanye isalathisi kwiSebe leNzululwazi yeKhompyutha kunye nobuNjineli, iMichigan State University. "I-Pattern Recognition and Image Processing (PRIP) iFakhalthi yeLebhu kunye nabafundi baphanda ngokusetyenziswa koomatshini ukuqonda iipateni okanye izinto. Iindlela ziphuhliswa ukuqonda izinto, ukufumanisa ukuba zeziphi iimpawu zazo ezahlula kwabanye, kunye nokuyila ii-algorithms ezinokuthi zisetyenziswe ngumatshini ukwenza ulwahlulo. Izicelo ezibalulekileyo zibandakanya ukuqaphela ubuso, ukuchongwa kweminwe, ukuhlalutya komfanekiso woxwebhu, ukwakhiwa kwemodeli ye-3D, ukukhangela i-robot, kunye nokubonwa / ukuhlola idatha ye-volumetric ye-3D. Iingxaki zophando lwangoku zibandakanya ukuqinisekiswa kwebhayometriki, ukucupha ngokuzenzekelayo kunye nokulandelela, i-HCI engenazandla, imodeli yobuso, i-watermarking yedijithali kunye nokuhlalutya isakhiwo samaxwebhu e-intanethi. Abasandula ukuthweswa izidanga kule lebhu basebenzele ukuqatshelwa kokubhala ngesandla, ukuqinisekiswa komsayino, ukufunda okubonwayo, kunye nokufunyanwa kwemifanekiso.”

⦁ UkuQatshelwa koBuso
Siyazi ukuba iinkqubo zokuqaphela ubuso ziya ngokuya zithandwa njengeendlela zokukhupha ulwazi lwebhayometriki. Ukuqondwa kobuso kunendima ebalulekileyo kwiinkqubo zebhayometriki kwaye inomtsalane kwizicelo ezininzi ezibandakanya ukucupha okubonakalayo kunye nokhuseleko. Ngenxa yokwamkelwa koluntu ngokubanzi kwemifanekiso yobuso kumaxwebhu ahlukeneyo, ukuqondwa kobuso kunamandla amakhulu okuba sisizukulwana esilandelayo sebhayometriki yeteknoloji yokuzikhethela.

IiNkqubo zokuQaphela umfanekiso

⦁ Uhlalutyo lwentshukumo
Imisebenzi emininzi inxulumene noqikelelo lwentshukumo apho ulandelelwano lomfanekiso lusetyenzwa khona ukuze kuveliswe uqikelelo lwesantya mhlawumbi kwindawo nganye emfanekisweni okanye kumboniso we-3D, okanye nakwikhamera evelisa imifanekiso. Imizekelo yaloo misebenzi yile:

⦁  Intshukumo ye-Ego
Ukumisela intshukumo engqongqo ye-3D (ujikelezo noguqulo) lwekhamera ukusuka kulandelelwano lomfanekiso oveliswe yikhamera.

⦁ Ukulandelela
Ukulandelela kulandela iintshukumo (eziqhelekile) ezincinci zamanqaku okanye izinto (umzekelo, izithuthi okanye abantu) kulandelelwano lwemifanekiso.

⦁ Ukuhamba okubonakalayo
Oku kukugqiba, kwindawo nganye emfanekisweni, ukuba loo ngongoma ihamba njani xa ithelekiswa nomfanekiso wendiza, oko kukuthi, intshukumo yayo ebonakalayo. Esi sindululo sisiphumo zombini indlela inqaku elihambelanayo le-3D elihamba ngayo kwindawo kunye nendlela ikhamera ehamba ngayo ngokumalunga nendawo.

⦁ Ukwakhiwa ngokutsha kwendawo
Kunikelwe umfanekiso omnye okanye (ngokwesiqhelo) ngaphezulu komboniso, okanye ividiyo, ulwakhiwo ngokutsha lwendawo lujolise ekhompyutheni imodeli ye-3D yendawo. Kwimeko elula imodeli ingaba yisethi yamanqaku e-3D. Iindlela eziphucukileyo ngakumbi zivelisa imodeli yomgangatho we-3D epheleleyo

⦁ Ukubuyiselwa komfanekiso
Injongo yokubuyisela umfanekiso kukususa ingxolo (ingxolo yenzwa, i-motion blur, njl.) kwimifanekiso. Eyona ndlela ilula enokwenzeka yokususa ingxolo ziindidi ezahlukeneyo zezihluzi ezifana nezihluzo ezidlulayo ezisezantsi okanye izihluzo eziphakathi. Iindlela ezinobugocigoci zithatha imodeli yendlela izakhiwo zomfanekiso wendawo ezibukeka ngayo, imodeli eyahlula kwingxolo. Ngokuhlalutya kuqala idatha yomfanekiso ngokwezakhiwo zemifanekiso yendawo, njengemigca okanye imiphetho, kwaye emva koko ulawule ukuhluzwa okusekelwe kulwazi lwendawo ukusuka kwinqanaba lokuhlalutya, umgangatho ongcono wokususa ingxolo udla ngokufunyanwa xa kuthelekiswa neendlela ezilula. Umzekelo kulo mmandla ngumzobo wabo. Ezinye iinkqubo zizimele zodwa ezisombulula umlinganiselo othile okanye ingxaki yobhaqo, ngelixa ezinye zenza inkqutyana yoyilo olukhulu oluthi, umzekelo, luqulethe iinkqubo ezisezantsi zolawulo lwezixhobo zoomatshini, ucwangciso, ugcino lweenkcukacha, umntu- ujongano lomatshini, njl.njl. Ukuphunyezwa okuthe ngqo kwenkqubo yombono wekhompyuter kukwaxhomekeke ekubeni ukusebenza kwayo kuchazwe kwangaphambili okanye ukuba inxalenye yayo inokufundwa okanye iguqulwe ngexesha lokusebenza. Kukho, nangona kunjalo, imisebenzi eqhelekileyo efumaneka kwiinkqubo ezininzi zombono wekhompyuter.

 

Ukufunda nzulu kunye nokuqatshelwa kwemifanekiso

Ukuqondwa komfanekiso kwakukho ngaphambi kwe-AI. Nangona kunjalo into yokufunda koomatshini iguqula iindlela zokuchonga into okanye ubuso bomntu. Ukufunda ngomatshini kusebenza kuphela xa kukho idatha yokondla, nangona kunjalo. Kuyo yonke i-automation ye-AI, ukuyinika umsebenzi wokuchonga imifanekiso ayisosicelo esilula. Ukuqonda kwethu okubonakalayo kuyindalo yesibini; yinto esiyicwangciselwe ukuba siyenze kwasebutsheni. Ukubuza okufanayo kumatshini akuyonkqubo ethe ngqo. Ngeso sizathu, enye yeendlela ezaziwa kakhulu zokuqatshelwa kwe-AI yi-convolutional neural networks (CNN). I-CNN yindlela egxininisa kwiipikseli ezibekwe ecaleni kwenye. Imifanekiso ekufutshane ithanda ukunxulumana, okuthetha ukuba into okanye ubuso buhambelana nomfanekiso ocacileyo ngakumbi.
Ngelixa iibhrendi zijonge ukwenza imali kwimithombo yeendaba zentlalo nangona ukuqondwa komfanekiso we-AI kuthwala izibonelelo ezicacileyo, iimeko zokusetyenziswa zibaleka nzulu kakhulu. Iimoto eziziqhubayo sele ziza kuba yinto enkulu elandelayo kwihlabathi leemoto, kwaye itekhnoloji yokuqaphela umfanekiso we-AI inceda ukubanika amandla. Imoto eziqhubayo ekwaziyo ukubona izinto nabantu endleleni ukuze ingatshayisa kubo ayizenzekeli. Kufuneka iqonde imifanekiso ukwenza izigqibo ezizizo. Imoto nganye eziqhubayo ifakelwe izivamvo ezininzi ukuze ikwazi ukuchonga ezinye izithuthi ezihambayo, abakhweli beebhayisikile, abantu - ngokusisiseko nantoni na enokuba yingozi. Imoto ezenzekelayo kufuneka ijongane neengozi zendlela ngendlela efanayo nomqhubi onamava. Kusekho imiba embalwa yokuayina phambi kokuba iimoto eziziqhubayo zifike endleleni ngo-2020. Kodwa xa i-automation yemoto ingena, ukuqondwa komfanekiso we-AI kuya kuba ngomnye wabaqhubi abakhulu emva kwabo basebenza ngokukhuselekileyo.
⦁ Ukufumana umfanekiso
Umfanekiso wedijithali uveliswa ngooluvo omnye okanye abaninzi bemifanekiso, leyo, ngaphandle kweentlobo ezahlukeneyo zeekhamera ezingevani nokukhanya, zibandakanya abenzi boluvo boluhlu, izixhobo zetomography, irada, iikhamera ze-ultra-sonic, njl. Kuxhomekeke kuhlobo lwesivamvo, isiphumo sedatha yomfanekiso. ngumfanekiso we-2D oqhelekileyo, umthamo we-3D, okanye ulandelelwano lomfanekiso. Amaxabiso epikseli ngokwesiqhelo angqinelana nokuqina kokukhanya kwibhendi enye okanye aliqela okubonwayo (imifanekiso engwevu okanye imifanekiso enemibala), kodwa inokunxulunyaniswa nemilinganiselo eyahlukeneyo yomzimba, enjengobunzulu, ukufunxwa okanye ukubonakaliswa kwamaza e-sonic okanye i-electromagnetic, okanye iresonance yenyukliya.
⦁ Ukusetyenzwa kwangaphambili:
Ngaphambi kokuba indlela yombono wekhompyutheni isetyenziswe kwidatha yomfanekiso ukuze kukhutshwe isicatshulwa esithile solwazi, ngokuqhelekileyo kuyimfuneko ukucutshungulwa kwedatha ukuze kuqinisekiswe ukuba iyanelisa iingcamango ezithile ezichazwe yindlela. Imizekelo yile
1. Ukuphinda kwenziwe iisampulu ukuze kuqinisekiswe ukuba inkqubo yolungelelwaniso lomfanekiso ichanekile.
2. Ukunciphisa ingxolo ukwenzela ukuqinisekisa ukuba ingxolo yenzwa ayizisi ulwazi lobuxoki.
3. Ukwandiswa kothelekiso ukuqinisekisa ukuba ulwazi olufanelekileyo luyabhaqwa.
4. Ukubonakaliswa kwesithuba sokulinganisa ukunyusa ukwakheka kwemifanekiso kwizikali ezifanelekileyo zendawo.
⦁ Ukutsalwa kophawu:
Iimpawu zemifanekiso kumanqanaba ahlukeneyo obunzima zicatshulwa kwidatha yomfanekiso. Imizekelo eqhelekileyo yeempawu ezinjalo ziyimigca, imiphetho kunye neengqungquthela
Amanqaku omdla asekuhlaleni afana neekona, iiblobhu okanye amanqaku. Iimpawu ezinzima ngakumbi zinokunxulumana nokuthungwa, imilo okanye intshukumo.
⦁ Ukufunyanwa/ukwahlulwa:
Ngexesha elithile ekuqhubeni isigqibo senziwa malunga nokuba yeyiphi imimandla yomfanekiso okanye imimandla yomfanekiso echaphazelekayo ekuqhubeni phambili. Imizekelo yile
1. Ukukhethwa kweseti ethile yamanqaku omdla
2. Ukwahlulwa kwemimandla yomfanekiso omnye okanye emininzi equlethe into ethile enomdla.
⦁ Ukusetyenzwa kwenqanaba eliphezulu:
Kweli nyathelo igalelo ngokuqhelekileyo liseti encinci yedatha, umzekelo iseti yamanqaku okanye ummandla we-animage ekucingelwa ukuba uqulethe into ethile. Ukuqhubekeka okuseleyo kusebenza, umzekelo:
1. Ukuqinisekisa ukuba idatha iyanelisa imodeli-based kunye neenkcukacha zesicelo.
2. Uqikelelo lweeparamitha ezithile zesicelo, ezifana nokuma kwento okanye ubukhulu bezinto.
3. Ukuhlela into efunyenweyo kwiindidi ezahlukeneyo.Ngoko, ukulungiswa komfanekiso kunceda i-AI ukuchonga umfanekiso kwaye uphendule ngokuchongwa komfanekiso.

Ikamva elingenamthungo lemifanekiso

Njengoko itekhnoloji iphucuka, ukuqondwa komfanekiso kuya kubuya neziphumo ezingaphezulu. Intloko yoFundo loMatshini eLobster, uVladimir Pavlov uthi, "Isiseko semathematika sokuqondwa kwezinto sele sikhona ixesha elide, kodwa amathuba obuchwepheshe bokusebenzisa i-algorithms yombono wekhompyuter avele mva nje. Sele, iinethiwekhi ze-neural zivumela ukwenza ii-detectors ezigqibeleleyo ezikwaziyo ukusebenza ngcono kunabantu. I-jerk enkulu ibamba ubukho beesethi zedatha eziphawulweyo zoqeqesho, kodwa kwixesha elizayo, oku akuyi kuba yingxaki. Iinjineli zembono yeKhompyutha zisebenza ngokukhutheleyo kwii-algorithms zokuzifundela ".Ngekamva elichatshazelwa kakhulu lunxibelelwano olubonakalayo, ukuqondwa kwemifanekiso kuya kuba yinto ephambili emva kwemifanekiso emininzi esiyibonayo. Zombini kubomi bokwenyani nakwi-intanethi.