Ubuchwepheshe namamodeli esistimu yethu yokubona yekhompyutha yesikhathi esizayo adalwe futhi athuthukiswa kancane kancane nakumaphrojekthi ahlukene enkampani yethu - ku-Mail, Cloud, Search. Bavuthwa njengoshizi omuhle noma i-cognac. Ngolunye usuku saqaphela ukuthi amanethiwekhi ethu e-neural abonisa imiphumela emihle kakhulu ekuqashelweni, futhi sanquma ukuwahlanganisa abe umkhiqizo owodwa we-b2b - Umbono - manje esiwusebenzisa ngokwethu futhi esikunikeza ukuthi uwusebenzise.

Namuhla, ubuchwepheshe bethu bokubona ngekhompyutha ku-platform ye-Mail.Ru Cloud Solutions busebenza ngempumelelo futhi buxazulula izinkinga ezingokoqobo eziyinkimbinkimbi kakhulu. Isekelwe enanini lamanethiwekhi e-neural aqeqeshwe kumasethi wethu wedatha futhi asebenza ngokukhethekile ekuxazululeni izinkinga ezisetshenzisiwe. Zonke izinsiza zisebenza ezindaweni zethu zeseva. Ungakwazi ukuhlanganisa i-Vision API yomphakathi ezinhlelweni zakho zokusebenza, lapho wonke amakhono esevisi etholakala khona. I-API iyashesha - sibonga amaseva we-GPU, isikhathi sokuphendula esimaphakathi ngaphakathi kwenethiwekhi yethu ngu-100 ms.

Iya ekati, kunendaba enemininingwane kanye nezibonelo eziningi zomsebenzi we-Vision.

Isibonelo sesevisi lapho thina ngokwethu sisebenzisa khona ubuchwepheshe obushiwo bokubona ubuso Izenzakalo. Enye yezingxenye zayo yi-Vision photo stands, esiyifaka ezingqungqutheleni ezihlukahlukene. Uma usondela endaweni yokuma yezithombe, thatha isithombe ngekhamera eyakhelwe ngaphakathi bese ufaka i-imeyili yakho, uhlelo luzothola ngokushesha phakathi kohlu lwezithombe lezo othwetshulwe kuzo abathwebuli bezithombe zengqungquthela, futhi, uma uthanda, izokuthumela izithombe ezitholakele nge-imeyili. Futhi asikhulumi ngezithombe ezithathwe esiteji—Umbono ukuqaphela ngisho nangemuva esixukwini sezivakashi. Vele, akuzona izithombe ezizimele ngokwazo ezibonwayo, lawa amathebulethi asezindaweni ezinhle avele athathe izithombe zezivakashi ngamakhamera azo akhelwe ngaphakathi futhi adlulisele imininingwane kumaseva, lapho kwenzeka khona wonke umlingo wokuqashelwa. Futhi sibone izikhathi ezingaphezu kwesisodwa ukuthi kumangalisa kanjani ukusebenza kahle kobuchwepheshe ngisho naphakathi kochwepheshe bokubona izithombe. Ngezansi sizokhuluma ngezibonelo ezithile.

1. Imodeli yethu Yokuqaphela Ubuso

1.1. Inethiwekhi ye-Neural nesivinini sokucubungula

Ukuze sibonwe, sisebenzisa ukuguqulwa kwemodeli yenethiwekhi ye-neural ye-ResNet 101. Ukuhlanganisa Okumaphakathi ekugcineni kushintshaniswa isendlalelo esixhumeke ngokugcwele, esifana nendlela okwenziwa ngayo ku-ArcFace. Kodwa-ke, ubukhulu bezethulo ze-vector buyi-128, hhayi 512. Isethi yethu yokuqeqeshwa iqukethe cishe izithombe eziyizigidi ezingu-10 zabantu abangu-273.

Imodeli isebenza ngokushesha kakhulu ngenxa yesakhiwo esikhethwe ngokucophelela sokumisa iseva kanye nekhompyutha ye-GPU. Kuthatha kusukela ku-100 ms ukuthola impendulo evela ku-API kumanethiwekhi ethu angaphakathi - lokhu kufaka phakathi ukubona ubuso (ukuthola ubuso esithombeni), ukubona nokubuyisela i-PersonID kumpendulo ye-API. Ngomthamo omkhulu wedatha engenayo - izithombe namavidiyo - kuzothatha isikhathi esengeziwe ukudlulisa idatha kusevisi nokuthola impendulo.

1.2. Ukuhlola ukusebenza kwemodeli

Kodwa ukunquma ukusebenza kahle kwamanethiwekhi e-neural kuwumsebenzi ongaqondakali kakhulu. Izinga lomsebenzi wabo lincike ekutheni amamodeli aqeqeshelwe amaphi amasethi edatha nokuthi athuthukiselwe ukusebenza ngedatha ethile.

Siqale ukuhlola ukunemba kwemodeli yethu ngokuhlolwa kokuqinisekisa kwe-LFW okudumile, kodwa kuncane kakhulu futhi kulula. Ngemuva kokufinyelela ukunemba okungu-99,8%, akusasizi. Kunomncintiswano omuhle wokuhlola amamodeli wokuqashelwa - i-Megaface, lapho kancane kancane safinyelela ku-82% isikhundla 1. Ukuhlolwa kwe-Megaface kuqukethe izithombe eziyisigidi - iziphazamisi - futhi imodeli kufanele ikwazi ukuhlukanisa kahle izithombe eziyizinkulungwane ezimbalwa zosaziwayo ezivela ku-Facescrub. Idathasethi evela kuziphazamisi. Kodwa-ke, ngemva kokusula amaphutha okuhlolwa kwe-Megaface, sithole ukuthi ngenguqulo esuliwe sifinyelela ukunemba kwe-98% izinga 1 (izithombe zosaziwayo ngokuvamile zicacile). Ngakho-ke, bakha isivivinyo sokuhlonza esihlukile, esifana ne-Megaface, kodwa ngezithombe zabantu "abavamile". Sabe sesithuthukisa ukunemba kokuqashelwa kumadathasethi ethu futhi saqhubekela phambili. Ngaphezu kwalokho, sisebenzisa ukuhlolwa kwekhwalithi yokuhlanganisa okuqukethe izithombe eziyizinkulungwane ezimbalwa; ilingisa ukumaka ubuso efwini lomsebenzisi. Kulokhu, amaqoqo angamaqembu abantu abafanayo, iqembu elilodwa lomuntu ngamunye obonakalayo. Sihlole ikhwalithi yomsebenzi kumaqembu wangempela (iqiniso).

Yiqiniso, amaphutha okuqaphela ayenzeka nganoma iyiphi imodeli. Kodwa izimo ezinjalo zivame ukuxazululwa ngokulungisa kahle imingcele yemibandela ethile (kuzo zonke izingqungquthela sisebenzisa ama-threshold afanayo, kodwa, isibonelo, kumasistimu okulawula ukufinyelela kufanele sandise kakhulu imingcele ukuze kube nemibono embalwa yamanga). Iningi labavakashi benkomfa labonwa ngendlela efanele amadokodo ethu ezithombe ze-Vision. Kwesinye isikhathi othile wayebheka ukubuka kuqala okusikiwe bese ethi, “Isistimu yakho yenze iphutha, akumina.” Sabe sesivula isithombe sonke, kwavela ukuthi ngempela kwakukhona lesi sivakashi esithombeni, kuphela sasingamthwebuli, kodwa omunye umuntu, umuntu uvele engemuva endaweni efiphele. Ngaphezu kwalokho, inethiwekhi ye-neural ivamise ukubona ngendlela efanele ngisho nalapho ingxenye yobuso ingabonakali, noma umuntu emi kuphrofayela, noma ejike kancane. Uhlelo lungakwazi ukubona umuntu ngisho noma ubuso busendaweni yokuhlanekezelwa kwe-optical, ake sithi, lapho udubula nge-lens ebanzi.

1.3. Izibonelo zokuhlola ezimweni ezinzima

Ngezansi kunezibonelo zendlela inethiwekhi yethu ye-neural esebenza ngayo. Izithombe zihanjiswa kokokufakayo, okumele azilebule esebenzisa i-PersonID - isihlonzi esiyingqayizivele somuntu. Uma izithombe ezimbili noma ngaphezulu zine-ID efanayo, ngokusho kwamamodeli, lezi zithombe zibonisa umuntu ofanayo.

Masiqaphele ngokushesha ukuthi lapho sihlola, sinokufinyelela kumapharamitha ahlukahlukene kanye nemikhawulo yemodeli esingayilungisa ukuze sizuze umphumela othile. I-API yomphakathi ithuthukiselwe ukunemba okuphezulu ezimweni ezivamile.

Ake siqale ngento elula kakhulu, ngokubona ubuso obubheke phambili.

Nokho, lokho kwakulula kakhulu. Masixabanise umsebenzi, sengeze intshebe kanye nedlanzana leminyaka.

Abanye bazothi lokhu kwakungenzima kakhulu, ngoba kuzo zombili izimo ubuso bonke bubonakala, futhi ulwazi oluningi mayelana nobuso luyatholakala ku-algorithm. Kulungile, asiguqule u-Tom Hardy abe yiphrofayela. Le nkinga iyinkimbinkimbi kakhulu, futhi sachitha umzamo omkhulu ukuyixazulula ngempumelelo ngenkathi sigcina izinga eliphansi lamaphutha: sikhethe isethi yokuqeqeshwa, sacabanga ngokwakhiwa kwenethiwekhi ye-neural, sacija imisebenzi yokulahlekelwa futhi sathuthukisa ukucubungula kwangaphambili. wezithombe.

Masimgqokise isigqoko:

Ngendlela, lesi yisibonelo sesimo esinzima ikakhulukazi, njengoba ubuso bufihliwe kakhulu, futhi esithombeni esingezansi kukhona nomthunzi ojulile ofihla amehlo. Ekuphileni kwangempela, abantu bavame ukushintsha ukubukeka kwabo ngosizo lwezibuko ezimnyama. Asenze okufanayo ngoTom.

Kulungile, ake sizame ukuphonsa izithombe ezivela eminyakeni eyahlukene, futhi kulokhu sizozama ngomlingisi ohlukile. Ake sithathe isibonelo esiyinkimbinkimbi kakhulu, lapho izinguquko ezihlobene nobudala zigqama kakhulu. Isimo asikho kude; kwenzeka kaningi lapho udinga ukuqhathanisa isithombe esisepasipoti nobuso bomphathi. Ngemuva kwakho konke, isithombe sokuqala sengezwe epasipoti lapho umnikazi eneminyaka engu-20 ubudala, futhi lapho eneminyaka engu-45 umuntu angashintsha kakhulu:

Ucabanga ukuthi uchwepheshe oyinhloko ohambweni olungenakwenzeka akakashintshi kakhulu ngokuya ngeminyaka? Ngicabanga ukuthi ngisho nabantu abambalwa bangahlanganisa izithombe eziphezulu nezingezansi, umfana ushintshe kakhulu eminyakeni edlule.

Amanethiwekhi e-Neural ahlangabezana nezinguquko ekubukekeni kaningi. Isibonelo, ngezinye izikhathi abesifazane bangashintsha kakhulu isithombe sabo ngosizo lwezimonyo:

Manje ake sihlanganise lo msebenzi nakakhulu: ake sithi izingxenye ezihlukene zobuso zimbozwe ezithombeni ezahlukene. Ezimweni ezinjalo, i-algorithm ayikwazi ukuqhathanisa amasampuli wonke. Nokho, i-Vision iziphatha kahle izimo ezinjengalezi.

Ngendlela, kungaba nobuso obuningi esithombeni; isibonelo, abantu abangaphezu kwe-100 bangangena esithombeni esijwayelekile sehholo. Lesi isimo esinzima samanethiwekhi we-neural, njengoba ubuso obuningi bungakhanyiswa ngendlela ehlukile, obunye ngaphandle kokugxila. Nokho, uma isithombe sithathwe ngokulungiswa okwanele kanye nekhwalithi (okungenani amaphikseli angu-75 isikwele ngasinye esimboze ubuso), i-Vision izokwazi ukusibona futhi isibone.

Okukhethekile kwezithombe ezibikwayo nezithombe ezivela kumakhamera agadayo ukuthi abantu bavame ukufiphala ngenxa yokuthi bebengagxilile noma bebehamba ngaleso sikhathi:

Futhi, amandla okukhanyisa angahluka kakhulu kusuka esithombeni kuye esithombeni. Nalokhu, nakho, kuvame ukuba yisikhubekiso; ama-algorithms amaningi anobunzima obukhulu ekucubunguleni izithombe ezimnyama kakhulu nezilula kakhulu, ingasaphathwa eyokumadanisa ngokunembile. Ake ngikukhumbuze ukuthi ukuze uzuze lo mphumela udinga ukulungisa ama-threshold ngendlela ethile; lesi sici asikatholakali esidlangalaleni. Sisebenzisa inethiwekhi efanayo ye-neural kuwo wonke amakhasimende; inezindodla ezifanele imisebenzi eminingi ebonakalayo.

Sisanda kwethula inguqulo entsha yemodeli ebona ubuso base-Asia ngokunemba okuphezulu. Lokhu kwakuke kwaba yinkinga enkulu, eyayibizwa nangokuthi “ukufunda ngomshini” (noma “inethiwekhi ye-neural”) ukucwasa ngokwebala. Amanethiwekhi emizwa aseYurophu nawaseMelika abubona kahle ubuso baseCaucasus, kodwa ngobuso bukaMongoloid noNegroid isimo sasibi kakhulu. Mhlawumbe, eChina isimo sasihluke ngokuphelele. Konke kumayelana nokuqeqeshwa kwamasethi edatha abonisa izinhlobo ezivelele zabantu ezweni elithile. Nokho, isimo siyashintsha, namuhla le nkinga ayinzima kangako. Umbono awunankinga nabantu bezinhlanga ezahlukene.

Ukuqashelwa kobuso kungenye yezinhlelo eziningi zobuchwepheshe bethu; Umbono ungaqeqeshelwa ukubona noma yini. Isibonelo, amapuleti elayisensi, okuhlanganisa nezimo ezinzima kuma-algorithms: kuma-engeli abukhali, angcolile futhi anzima ukufunda amapuleti elayisensi.

2. Amacala okusetshenziswa okungokoqobo

2.1. Ukulawula ukufinyelela ngokomzimba: lapho abantu ababili besebenzisa iphasi efanayo

Ngosizo lwe-Vision, ungasebenzisa izinhlelo zokuqopha ukufika nokuhamba kwabasebenzi. Uhlelo lwendabuko olusekelwe ekudluleni kwe-elekthronikhi lunobubi obusobala, isibonelo, ungadlula abantu ababili usebenzisa ibheji eyodwa. Uma uhlelo lokulawula ukufinyelela (ACS) lulekelelwa nge-Vision, luzorekhoda ngokwethembeka ukuthi ubani oze/ohambile futhi nini.

2.2. Ukulandelela isikhathi

Leli cala lokusebenzisa le-Vision lihlobene eduze nedlule. Uma wengeza isistimu yokufinyelela ngesevisi yethu yokuqaphela ubuso, ngeke ikwazi nje ukubona ukwephulwa kokulawula ukufinyelela, kodwa futhi nokubhalisa ubukhona bangempela babasebenzi esakhiweni noma esikhungweni. Ngamanye amazwi, i-Vision izokusiza ukuthi ucabangele ngobuqotho ukuthi ubani ofike emsebenzini futhi wahamba ngasiphi isikhathi, nokuthi ubani oweqa umsebenzi ngokuphelele, noma ngabe ozakwabo bamkhava phambi kwabaphathi bakhe.

2.3. Izibalo Zevidiyo: Ukulandelela Abantu Nokuphepha

Ngokulandela abantu usebenzisa i-Vision, ungakwazi ukuhlola ngokunembile ithrafikhi yangempela yezindawo zokuthenga, iziteshi zesitimela, amaphaseji, imigwaqo kanye nezinye izindawo eziningi zomphakathi. Ukulandelela kwethu kungase futhi kube usizo olukhulu ekulawuleni ukufinyelela, isibonelo, endaweni yokugcina izimpahla noma kwezinye izakhiwo zehhovisi ezibalulekile. Futhi-ke, ukulandelela abantu nobuso kusiza ukuxazulula izinkinga zokuphepha. Uke wabanjwa umuntu oweba esitolo sakho? Engeza i-PersonID yakhe, ebuyiselwe ngu-Vision, ohlwini oluvinjelwe lwesofthiwe yakho yokuhlaziya ividiyo, futhi ngokuzayo isistimu izokwazisa ngokushesha ukuphepha uma lolu hlobo luvela futhi.

2.4. Kwezohwebo

Amabhizinisi okuthengisa namasevisi ahlukahlukene anentshisekelo ekwazisweni komugqa. Ngosizo lwe-Vision, ungabona ukuthi lesi akusona isixuku sabantu esingahleliwe, kodwa umugqa, futhi unqume ubude baso. Bese uhlelo lwazisa labo abaphethe ulayini ukuze bakwazi ukuthola isimo: kungaba khona ukuthutheleka kwezivakashi futhi kudingeka kubizwe abasebenzi abengeziwe, noma othile uyaxega emsebenzini wakhe.

Omunye umsebenzi othokozisayo ukuhlukanisa abasebenzi benkampani ehholo nezivakashi. Ngokuvamile, uhlelo luqeqeshelwe ukuhlukanisa izinto ezigqoke izingubo ezithile (ikhodi yokugqoka) noma ngesici esithile esihlukile (isikhafu esinophawu, ibheji esifubeni, njalonjalo). Lokhu kusiza ekuhloleni ngokunembe kakhudlwana inani lababekhona (ukuze abasebenzi “bangakhuphulisi” izibalo zabantu ehholo ngokuba khona nje kwabo).

Usebenzisa ukuqashelwa kobuso, ungaphinda uhlole izethameli zakho: buyini ukwethembeka kwezivakashi, okungukuthi, bangaki abantu ababuyela endaweni yakho nokuthi bangaki imvamisa. Bala ukuthi zingaki izivakashi ezihlukile eziza kuwe ngenyanga. Ukuze uthuthukise izindleko zokuheha nokugcinwa, ungathola futhi ushintsho kuthrafikhi ngokuya ngosuku lweviki kanye nesikhathi sosuku.

Ama-Franchisor kanye nezinkampani zamaketanga zinga-oda ukuhlolwa kwesithombe sekhwalithi yokuthengisa ezitolo ezahlukahlukene: ukuba khona kwama-logo, izimpawu, amaphosta, amabhanela, nokunye.

2.5. Ngezokuthutha

Esinye isibonelo sokuqinisekisa ukuphepha kusetshenziswa izibalo zevidiyo ukuhlonza izinto ezilahliwe emahholo ezikhumulo zezindiza noma eziteshini zezitimela. Umbono ungaqeqeshelwa ukuqaphela izinto zamakhulu amakilasi: izingcezu zefenisha, izikhwama, amapotimende, izambulela, izinhlobo ezahlukene zezingubo, amabhodlela, njalonjalo. Uma isistimu yakho yokuhlaziya ividiyo ithola into engenamnikazi futhi iyibona kusetshenziswa i-Vision, ithumela isignali kusevisi yezokuvikela. Umsebenzi ofanayo uhlotshaniswa nokutholwa okuzenzakalelayo kwezimo ezingavamile ezindaweni zomphakathi: othile uzizwa egula, noma othile ubhema endaweni engafanele, noma umuntu uwela emigwaqweni, njalonjalo - wonke lawa maphethini angabonwa ngezinhlelo zokuhlaziya ividiyo. ngokusebenzisa Vision API.

2.6. Ukugeleza kwedokhumenti

Olunye uhlelo lwesikhathi esizayo oluthokozisayo lwe-Vision esiluthuthukisayo njengamanje ukuqashelwa kwemibhalo kanye nokwehlukaniswa kwakho okuzenzakalelayo kusizindalwazi. Esikhundleni sokufaka ngesandla (noma okubi nakakhulu, ukungena) uchungechunge olungapheli, izinombolo, izinsuku zokukhishwa, izinombolo ze-akhawunti, imininingwane yasebhange, izinsuku nezindawo zokuzalwa kanye neminye imininingwane eminingi esemthethweni, ungaskena amadokhumenti futhi uwathumele ngokuzenzakalelayo esiteshini esivikelekile nge- I-API eya efwini, lapho uhlelo luzobona khona le mibhalo ngokuhamba kwesikhathi, ihlukanise futhi ibuyisele impendulo enedatha ngefomethi edingekayo yokungena ngokuzenzakalelayo kusizindalwazi. Namuhla Umbono usuvele uyayazi indlela yokuhlukanisa imibhalo (kuhlanganise ne-PDF) - ihlukanisa phakathi kwamaphasipoti, i-SNILS, i-TIN, izitifiketi zokuzalwa, izitifiketi zomshado nezinye.

Yebo, inethiwekhi ye-neural ayikwazi ukubhekana nazo zonke lezi zimo ngaphandle kwebhokisi. Esimweni ngasinye, imodeli entsha yakhelwe ikhasimende elithile, izici eziningi, ama-nuances kanye nezidingo ziyacatshangelwa, amasethi wedatha akhethiwe, futhi ukuphindaphinda kokuqeqeshwa, ukuhlolwa, nokucushwa kuyenziwa.

3. Uhlelo lokusebenza lwe-API

“Isango lokungena” likaVision labasebenzisi i-REST API. Ingathola izithombe, amafayela evidiyo kanye nokusakaza okuvela kumakhamera enethiwekhi (ukusakazwa kwe-RTSP) njengokufakiwe.

Ukuze usebenzise i-Vision, udinga ubhalisele kusevisi ye-Mail.ru Cloud Solutions futhi uthole amathokheni okufinyelela (client_id + client_secret). Ukufakazela ubuqiniso komsebenzisi kwenziwa kusetshenziswa iphrothokholi ye-OAuth. Idatha yomthombo emizimbeni yezicelo ze-POST ithunyelwa ku-API. Futhi ekuphenduleni, iklayenti lithola ku-API umphumela wokuqashelwa ngefomethi ye-JSON, futhi impendulo yakhiwe: iqukethe ulwazi mayelana nezinto ezitholakele kanye nezixhumanisi zazo.

Impendulo eyisampula

{
   "status":200,
   "body":{
      "objects":[
         {
            "status":0,
            "name":"file_0"
         },
         {
            "status":0,
            "name":"file_2",
            "persons":[
               {
                  "tag":"person9"
                  "coord":[149,60,234,181],
                  "confidence":0.9999,
                  "awesomeness":0.45
               },
               {
                  "tag":"person10"
                  "coord":[159,70,224,171],
                  "confidence":0.9998,
                  "awesomeness":0.32
               }
            ]
         }

         {
            "status":0,
            "name":"file_3",
            "persons":[
               {
               "tag":"person11",
               "coord":[157,60,232,111],
               "aliases":["person12", "person13"]
               "confidence":0.9998,
               "awesomeness":0.32
               }
            ]
         },
         {
            "status":0,
            "name":"file_4",
            "persons":[
               {
               "tag":"undefined"
               "coord":[147,50,222,121],
               "confidence":0.9997,
               "awesomeness":0.26
               }
            ]
         }
      ],
      "aliases_changed":false
   },
   "htmlencoded":false,
   "last_modified":0
}

Impendulo iqukethe ubuhle bepharamitha ezithakazelisayo - lokhu "ukuphola" kobuso obunemibandela esithombeni, ngosizo lwayo sikhetha isithombe esihle kakhulu sobuso ngokulandelana. Siqeqeshe inethiwekhi ye-neural ukubikezela amathuba okuthi isithombe sizothandwa ezinkundleni zokuxhumana. Ukuba ngcono kwekhwalithi yesithombe kanye nobuso obumomotheka kakhulu, kukhulu ukumangala.

I-API Vision isebenzisa umqondo obizwa ngokuthi isikhala. Leli ithuluzi lokudala amasethi ahlukene obuso. Izibonelo zezikhala uhlu olumnyama nolumhlophe, uhlu lwezivakashi, abasebenzi, amaklayenti, njll. Kuthokheni ngayinye ku-Vision, ungakha izikhala ezifika kweziyi-10, isikhala ngasinye singaba nama-PersonID afinyelela ezinkulungwaneni ezingu-50, okungukuthi, kufika ezinkulungwaneni ezingu-500. ngethokheni . Ngaphezu kwalokho, inani lamathokheni nge-akhawunti ngayinye alinqunyelwe.

Namuhla i-API isekela izindlela ezilandelayo zokutholwa nezindlela zokuqaphela:

• Bona/Setha - ukutholwa nokubonwa kobuso. Yabela i-PersonID ngokuzenzakalelayo kumuntu ngamunye ohlukile, ibuyisela i-PersonID kanye nezixhumanisi zabantu abatholiwe.
• Susa - ukususa i-PersonID ethile kusizindalwazi somuntu.
• I-Truncate - isula sonke isikhala ku-PersonID, iwusizo uma isetshenziswe njengendawo yokuhlola futhi udinga ukusetha kabusha isizindalwazi ukuze sikhiqize.
• Thola - ukutholwa kwezinto, izigcawu, amapuleti elayisensi, izimpawu zendawo, olayini, njll. Ibuyisela ikilasi lezinto ezitholiwe kanye nezixhumanisi zazo
• Thola imibhalo - ithola izinhlobo ezithile zemibhalo ye-Russian Federation (ihlukanisa ipasipoti, i-SNILS, inombolo kamazisi yentela, njll.).

Futhi maduze nje siqedela umsebenzi wezindlela ze-OCR, ukunquma ubulili, ubudala kanye nemizwa, kanye nokuxazulula izinkinga zokuthengisa, okungukuthi, ukulawula ngokuzenzakalelayo ukuboniswa kwezimpahla ezitolo. Ungathola imibhalo ephelele ye-API lapha: https://mcs.mail.ru/help/vision-api

4. Isiphetho

Manje, nge-API yomphakathi, ungakwazi ukufinyelela ukubonakala kobuso ezithombeni nakumavidiyo; ukuhlonza izinto ezihlukahlukene, amapuleti elayisensi, izimpawu zendawo, imibhalo kanye nezigcawu zonke kusekelwa. Izimo Isicelo - ulwandle. Woza, uhlole insizakalo yethu, uyibeke njengemisebenzi enzima kakhulu. Izinkokhelo zokuqala ezingu-5000 zimahhala. Mhlawumbe kuzoba "isithako esingekho" samaphrojekthi akho.

Ungakwazi ukufinyelela ngokushesha i-API lapho ubhalisa futhi usuxhumekile. Umbono. Bonke abasebenzisi be-Habra bathola ikhodi yephromoshini yemisebenzi eyengeziwe. Ngicela ungibhalele i-imeyili oyisebenzisele ukubhalisa i-akhawunti yakho!

Source: www.habr.com