Kusukela ngoJulayi 3 kuya kuJulayi 16 eNizhny Novgorod State University. N.I. U-Lobachevsky ubambe i-Intel Interuniversity Summer School ku-Computer Vision - I-Computer Vision Summer Camp, lapho abafundi abangaphezu kwe-100 babambe iqhaza. Lesi sikole sasihloselwe abafundi bezobuchwepheshe abavela emanyuvesi aseNizhny Novgorod abathanda umbono wekhompyutha, ukufunda okujulile, amanethiwekhi we-neural, i-Intel OpenVINO, i-OpenCV.
Kulesi sihloko sizokwabelana ngokuthi ukukhethwa kweSikole kwenzeka kanjani, lokho abakufundayo, lokho abafundi abakwenzile engxenyeni engokoqobo, futhi sikhulume nangamanye amaphrojekthi ethulwe ekuziphenduleleni.
Inqubo yokukhetha kanye namafomu okubamba iqhaza
Sanquma ukunikeza izingane ithuba lokukhetha ukufaka isicelo sezinhlobo ezimbili zemfundo: isikhathi esigcwele kanye netoho. Ezifundweni zesikhashana nezesikhashana, abafundi abazange bakhethe futhi babhaliswe ngokushesha. Babehambela izinkulumo kuphela, phakathi nesonto, ekuseni. Izingane nazo zithole ithuba lokuqeda imisebenzi eyenziwayo bese ziyithumela ukuze kuhlolwe othisha.
Ukuze bafanelekele ukuhlolwa ngokugcwele, abafana kwadingeka beze ehhovisi le-Intel ukuze baxoxisane nekhomishini. Umehluko efomini letoho kanye neletoho wawuwukuthi, ngaphezu kwezinkulumo, abahlanganyeli basekamu benza imisebenzi eyenziwayo nabaqaphi - othisha be-UNN nonjiniyela abavela ku-Intel. Ngesonto lesibili, izabelo ezingokoqobo zaphela futhi amaphrojekthi aqala, lapho abahlanganyeli basebenza ngamaqembu abantu abangu-3.
Ngesikhathi kuqhubeka inhlolokhono abafundi babuzwa imibuzo ngezibalo nezinhlelo, baphinde banikwa inkinga obekumele ixazululwe khona manjalo. Kuyaphawuleka ukuthi ikhomishana yayihlanganisa onjiniyela be-software, onjiniyela be-algorithm, nothisha basenyuvesi. N.I. Lobachevsky, ngakho-ke inhlolokhono yaba nemikhakha eminingi futhi engavamile. Ngokombono womuntu obuza imibuzo, bekuthakazelisa ukuthola ulwazi oluyisisekelo lwabafundi lobuchwepheshe ngokuphathelene nombono wekhompyutha, ngakho izihloko ezifana ne-C++/STL, i-OOP, ama-algorithms ayisisekelo kanye nezakhiwo zedatha, i-algebra eqondile, ukuhlaziya izibalo, izibalo ezihlukene kanye okuningi kwabuzwa. Phakathi kwemisebenzi okwakuseqhulwini kwakuwukuthola ukuthi abafundi bacabangani. Ikhomishana ibiphinde ibe nentshisekelo yokuthi bafunde kuphi, yiluphi ulwazi ababenalo ngaphambi kwalesi sikole (isibonelo, umsebenzi wesayensi) nokuthi lungasetshenziswa kanjani ngqo emkhakheni wokubona ngekhompyutha.
Bangu-78 abafundi ababambe iqhaza ngokugcwele kulo mqhudelwano, kanti izikhala zabafundi ezigcwele zingama-24. Izibalo zabahlanganyeli kanye nomehluko obonakalayo phakathi kwamafomu agcwele kanye nesikhashana wokubamba iqhaza kungabonakala kuthebula elingezansi:
Benzeni abafana amaviki ama-2?
Abafundi bajwayelana nethiyori kanye nokuzijwayeza imisebenzi esemqoka yombono wekhompyutha: ukuhlukaniswa kwezithombe, ukutholwa kwezinto kanye nokulandelela kwazo. Ingxenye yesifundo sesihloko ngasinye ngokuvamile yayihlanganisa uhambo lomlando ekuthuthukisweni kwezindlela zakudala zokuxazulula izinkinga zombono wekhompyutha nezindlela zesimanje zokuxazulula kusetshenziswa ukufunda ngomshini namanethiwekhi emizwa. Ithiyori yalandelwa umkhuba, lapho abafundi belande khona amamodeli enethiwekhi ye-neural adumile futhi bawethula besebenzisa imojuli ye-DNN yelabhulali ye-OpenCV, bakha uhlelo lokusebenza olungokwezifiso.
Izethulo zazo zonke izinkulumo zazifakwe endaweni yokugcina umphakathi , ukuze abafundi bahlale bevula futhi babuke ulwazi oludingekayo, kuhlanganise nokuphuma kwesikole. Bekungenzeka ukuxhumana nabafundisi, othisha abazijwayezayo kanye nonjiniyela be-Intel bukhoma futhi ngengxoxo ku-Gitter. Isikhathi seviki lephrojekthi siphinde saba yimpumelelo: saqala ngoLwesithathu, okwenze kwaba nokwenzeka ukusebenzisa impelasonto ngokusebenziseka kalula ngaphandle kwezinkulumo, nokwenza ngcono izinqumo zeqembu. Abahlanganyeli abanomthwalo wemfanelo kakhulu bachithe ingxenye yoMgqibelo ehhovisi le-Intel, lapho baklonyeliswe ngohambo olungahleliwe ngosuku olufanayo.
Bekunjani ukuvikela amaphrojekthi?
Ithimba ngalinye lanikezwa imizuzu eyi-10 yokukhuluma ngalokho elikwenzile phakathi nomsebenzi kanye nalokho elize kukho. Ngemuva kwalesi sikhathi, imizuzu engu-5 yaqala, lapho onjiniyela benkampani babuza abafana imibuzo futhi banikeza amathiphu amancane angabasiza ukuba bathuthukise iphrojekthi yabo noma bavimbele amaphutha akhona esikhathini esizayo. Ngamunye wabafana wazama ngokwabo njengesikhulumi, ebonisa ulwazi lwabo emkhakheni wombono wekhompyutha futhi eqinisekisa iqhaza labo ekudalweni kwephrojekthi, eyasisiza ukuba sicabangele futhi sifinyelele isiphetho mayelana nomhlanganyeli ngamunye esikoleni. Ukuzivikela kwenzeke ngaphezu kwamahora angu-3, kodwa sabanakekela abafana futhi sadambisa ukungezwani ngekhefu elifushane lekhofi, lapho abafana bengathatha khona umoya futhi baxoxe ngezinkinga nochwepheshe abaholayo be-Intel.
Ekupheleni kosuku, siklomelise indawo yokuqala, eyesibili yesibili neyesithathu. Kwakunzima kakhulu ukukhetha, ngoba iqembu ngalinye, iphrojekthi ngayinye yayinokunambitheka kwayo futhi yayihlukaniswa ngokuqala kwayo isethulo.
Abahlanganyeli besikhathi esigcwele be-CV Camp, ukuvikela iphrojekthi, ihhovisi le-Intel eNizhny Novgorod
Kwethulwa amaphrojekthi
Iglavu ehlakaniphile
Kusetshenziswa umtshina kanye ne-tracker kusetshenziswa i-OpenCV ukuzulazula okubonakalayo emkhathini. Ithimba lengeze ikhono lezinzwa ezijulile lisebenzisa amakhamera amabili. I-Microsoft Speech API isetshenziswa njengesixhumi esibonakalayo sokuphatha.
Isamukeli
Ukutholwa kokudla nokukhethwa kweresiphi yesitsha esenziwe ngomumo, kufaka phakathi izithako ezitholakele. Abafana babengawesabi umsebenzi futhi phakathi nesonto bamaka inombolo eyanele yezithombe bebodwa, baqeqesha umtshina besebenzisa i-TensorFlow Object Detection API futhi bangeza umqondo wokuthola iresiphi. Ilula futhi inambitheka!
Umhleli 2.0
Abahlanganyeli bephrojekthi basebenzise isethi yamanethiwekhi e-neural (ukusesha ubuso, ukujwayelekile kwesithombe sobuso ngamaphuzu abalulekile, ukubalwa kwesichazi sesithombe sobuso) ukuze kuqashelwe ubuso njengengxenye yomsebenzi wokusesha izingcezu kumavidiyo amade lapho umuntu othile ekhona. samanje. Uhlelo oluthuthukisiwe lungasetshenziswa njengohlelo losizo lokuhlelwa kwevidiyo, ukukhulula umuntu ekubukeni ividiyo ngokwakhe efuna izingcezu ezidingekayo. Ukusebenzisa amanethiwekhi e-neural kusuka , ithimba likwazile ukufeza isivinini esikhulu sohlelo lokusebenza: kwi-laptop ene-Intel Core i5 processor, isivinini sokucubungula ividiyo sasingamafreyimu angu-58 ngomzuzwana.
Ongaziwa
Ukudweba izibuko namamaski ebusweni bomuntu. Inethiwekhi ye-MTCNN isetshenziselwe ukuthola ubuso namaphoyinti abalulekile.
Ongaziwa
Omunye umsebenzi othokozisayo esihlokweni sokufihla ubuwena. Leli thimba lethule izinketho ezimbalwa zokuhlanekezela ubuso: ukufiphala kanye ne-pixelation. Ngesonto elilodwa, abafana abagcinanga nje ngokuthola umsebenzi, kodwa futhi banikeze imodi yokungaziwa komuntu othize (ngokuqashelwa kobuso).
Kulungele
Ithimba lephrojekthi "Yokufudumala" lixazulule inkinga yokudala umsizi wezemidlalo wokuzivocavoca kwekhanda. Futhi noma ngabe ukusetshenziswa kokugcina kwalolu hlelo lokusebenza kusenempikiswano, kwenziwa ucwaningo olunzulu ngokuqhathanisa ama-algorithms okuthola ubuso ahlukahlukene: ama-Haar cascades, amanethiwekhi avela ku-TensorFlow, OpenCV kanye ne-OpenVINO. Asifudumalanga ngokomzimba kuphela, kodwa nangokwengqondo!
Ngaphansi kuka-800
I-Nizhny Novgorod, idolobha lapho isikole senzeka khona, izoba neminyaka engu-2 eminyakeni emi-800, okusho ukuthi kunesikhathi esanele sokuqalisa iphrojekthi ethakazelisayo. Sicele izingane ukuthi zicabange ngomsebenzi wokudala umhlahlandlela, ngokusekelwe emfanekisweni we-facade yezakhiwo, onganikeza ulwazi mayelana nokuthi hlobo luni lwento ekhonjiswe esithombeni nokuthi yimaphi amaqiniso aziwayo ngakho. Ngokombono wethu, lo msebenzi wawungomunye onzima kakhulu, ngoba uhlobene nombono wekhompyutha we-classic, kodwa iqembu libonise umphumela ohloniphekile.
I-Rock Paper Scissors
Ngaphandle kwemikhawulo yesikhathi eqinile yokuqedela umsebenzi wokuklama, leli qembu alizange lesabe ukwenza isilingo ukuze liqeqeshe inethiwekhi yalo ye-neural ukuhlukanisa izindawo zezandla kugeyimu eyaziwayo.
Impendulo evela kubahlanganyeli
Sicele abafundi abavela ezifundweni ezahlukahlukene ukuthi babelane ngemibono yabo ngeSikole Sasehlobo:
Ngisanda kuba nenhlanhla yokuhambela i-Intel Computer Vision Summer Camp futhi kwaba yinto enhle kakhulu. Sithole ulwazi oluningi namakhono amasha emkhakheni we-CV, ukufakwa kwesoftware, ukulungisa amaphutha, saphinde sacwiliswa endaweni yokusebenza, sabhekana nezinkinga zangempela, saxoxa ngezixazululo ezingase zibe khona nozakwethu kanye nothisha besikole kuhlanganisa kuphela ukuxhumana nekhompuyutha. Nokho, akunjalo nhlobo. Umsebenzi wethu wokudala awunakuhlukaniswa nokuxhumana nabantu. Kwakungokuxhumana lapho umuntu ayengathola khona ulwazi oluyingqayizivele. Futhi ngayithanda le ngxenye yesikole kakhulu. Nokho, kukhona okushiyekayo okukodwa... ngemva kokuqeda izifundo zami ngangifuna ukuqhubeka! Ngaphezu kolwazi lwethiyori ku-DL namakhono asebenzayo ku-CV, ngithole umbono wokuthi yiziphi izindawo zezibalo okufanele zinikezwe ukunakwa okukhethekile nokuthi ibuphi ubuchwepheshe okufanele bufundwe. Ukuzinikela, ubungcweti nothando ngomsebenzi wabo wonjiniyela be-Intel nabacwaningi kube nomthelela ekukhetheni kwami isiqondiso ku-IT. Kungalokhu ngithanda ukubonga bonke abahleli besikole.
UKristina, unyaka wokuqala, HSE
Esikhathini esifushane kangako, isikole sikwazile ukuhlinzeka ngolwazi oluningi nokuzijwayeza ngesihloko sombono wekhompyutha. Futhi nakuba yayiklanyelwe ulwazi oluyisisekelo, izinkulumo zaziqukethe izinto eziningi zobuchwepheshe ofuna ukuziqonda futhi uchithe isikhathi esiningi ufunda. Abaqeqeshi nabafundisi bakulesi sikole baphendule ngokuzimisela yonke imibuzo futhi baxhumana nabafundi. Nokho, lapho ngiqeda umsebenzi wokugcina, kwadingeka ngingene ehlathini ukuze ngifake isicelo esiqediwe futhi ngihlangabezane nobunzima obungaveli njalo lapho ngifunda. Ithimba lethu lagcina lifake isicelo sokudlala umdlalo othi “rock-paper-scissors” ngekhompyutha. Siqeqeshe imodeli ukubona isibalo kukhamera yewebhu, sabhala okunengqondo futhi senza isixhumi esibonakalayo esisekelwe kuhlaka lwe-opencv. Isikole sihlinzeke ngokudla kokucabanga kanye ne-vector yokufunda nokuthuthukiswa okulandelayo. Ngijabula kakhulu ngokuthi ngibe yingxenye.
U-Sergey, unyaka wesi-3, i-UNN
Isikole asizange senze lokho engangikulindele. Izinkulumo zanikezwa abantu abanolwazi oluningi oluvela kubathuthukisi be-Intel. Ukuxhumana nabafundisi bekulokhu kuthakazelisa futhi kuwusizo, abeluleki bayasabela futhi bahlale bekulungele ukusiza Izinkulumo zimnandi ukuzilalela, izihloko zibalulekile futhi zifundisa. Kodwa ngase ngizazi ezinye izinto, futhi lezo engangingazazi azizange zisekelwe umkhuba nganoma iyiphi indlela, ngakho-ke indaba enhle ngempela ayizange iqondwe ngokugcwele futhi ifundwe yimina. Yebo, ulwazi oluningi luhlinzekelwe izinjongo zolwazi, ukuze ukwazi-ke ukulizama ekhaya, noma ube nombono wokuthi limayelana nani, kodwa bengisafuna ukusebenzisa amanye ama-algorithms akhona ngokwami ngaphansi kwe- ukugadwa kothisha abanolwazi abanganikeza iseluleko esihle noma usizo uma kukhona okwenzekayo kungasebenzi. Ngenxa yalokho, ekusebenzeni, izixazululo ezenziwe ngomumo zasetshenziswa, futhi ikhodi, omunye angase athi, yayibhalwe ngaphambili kithi; Amaphrojekthi abelula kakhulu, futhi uma uzama ukwenza umsebenzi ube nzima ngandlela thize, awubi naso isikhathi esanele sokuwenza ube sesimeni esizinzile, njengoba kwenzeka kithi.
Ngokuvamile, sonke isikole sibukeka njengohlobo oluthile lomdlalo ongewona obucayi kakhulu wonjiniyela, futhi lokhu kuyiphutha lengxenye ebonakalayo. Ngicabanga ukuthi kuyadingeka ukwandisa isikhathi esichithwa esikoleni, ukuhlanganisa izinto zokuzijwayeza ukuze ukwazi futhi kufanele uzibhale okuthile ngokwakho, into eyinkimbinkimbi ngempela futhi edingekayo, futhi ungasebenzisi esenziwe ngomumo, ukwenza umkhuba ube bushelelezi ekwandeni. ubunkimbinkimbi, izihloko zamaphrojekthi wemincintiswano kufanele zinikezwe ezinsukwini zokuqala, ukuze ukwaziswa okuvela ezifundweni kanye nemikhuba kusetshenziswe ngokushesha kumaphrojekthi akho futhi kuzoba nesikhathi esengeziwe sokuqalisa. Khona-ke isikhathi esichithwa esikoleni sizosebenza njengokuhlangenwe nakho okuhle kochwepheshe abafisayo.
UDmitry, unyaka wokuqala weziqu ze-master, NSTU
Isikole sasehlobo esivela ku-Intel sibe yithuba elihle lokuchitha leli hlobo wenza okuthandayo. Iqiniso lokuthi izinkulumo zanikezwa abasebenzi be-Intel ezihlobene nezinhlelo emkhakheni wombono wekhompyutha akuzange kungivumele ukuba ngiphumule; Zonke izinsuku zazidlula ngokushesha okukhulu, ngokungabonakali nangezithelo. Ithuba lokuqalisa iphrojekthi yami lingivumele ukuthi ngisebenze eqenjini elinabaqaphi abahle nabanye ababambiqhaza besikole. Lawa maviki amabili angachazwa kafushane kanje: athakazelisayo futhi adlulayo.
U-Elizaveta, unyaka wesi-2, i-UNN
Ekwindla (Okthoba-Novemba), uhlelo lwezemfundo lwe-Delta lulindelwe, ulwazi ongathola ngalo kithi . Hlala ubukele!
Source: www.habr.com