Ukususela ngoJulayi 3 ukuya kuJulayi 16 kwiYunivesithi yaseNizhny Novgorod State. N.I. U-Lobachevsky wabamba i-Intel Interuniversity Summer School kwi-Computer Vision-Computer Vision Summer Camp, apho abafundi abangaphezu kwe-100 bathatha inxaxheba. Esi sikolo sijolise kubafundi bezobugcisa abavela kwiiyunivesithi zaseNizhny Novgorod abanomdla kwimbono yekhompyutheni, ukufunda okunzulu, i-neural networks, i-Intel OpenVINO, i-OpenCV.
Kweli nqaku siza kwabelana ngendlela ukhetho lweSikolo lwenzeka ngayo, into abayifundayo, into eyenziwa ngabafundi kwinxalenye ebonakalayo, kwaye sithethe ngezinye zeeprojekthi ezinikezelwe ekukhuseleni.
Inkqubo yokhetho kunye neendlela zokuthatha inxaxheba
Sagqiba kwelokuba sinike abantwana ithuba lokuzikhethela ngokufaka izicelo zeendlela ezimbini zemfundo: ixesha elizeleyo nelomsebenzi wesingxungxo. Kwizifundo ezingesosigxina nezingesosigxina, abafundi abazange bakhethwe kwaye babhaliswa ngoko nangoko. Babesiya kwiintetho kuphela, phakathi evekini, kusasa. Abantwana bakwafumene ithuba lokugqibezela imisebenzi eyenziwayo baze bayithumele kuyo kuvavanywa ngootitshala.
Ukuze bafanelekele uviwo olusisigxina, abafana kwafuneka beze kwiofisi ye-Intel kudliwano-ndlebe kunye nekhomishini. Umahluko kwifom yethutyana kunye neyexeshana yayikukuba, ukongeza kwiintetho, abathathi-nxaxheba beenkampu baye benza imisebenzi yepraktikhali kunye nabagcini - ootitshala be-UNN kunye neenjineli ezivela kwi-Intel. Kwiveki yesibini, izabelo ezisebenzayo zaphela kwaye iiprojekthi zaqala, apho abathathi-nxaxheba basebenza ngamaqela abantu aba-3.
Ngexesha lodliwano-ndlebe abafundi babuzwa imibuzo malunga nezibalo neprogramu, kwaye banikwa nengxaki ekufuneka isonjululwe ngoko nangoko. Kuyaphawuleka ukuba ikhomishini ibibandakanya iinjineli zesoftware, iinjineli ze-algorithm, kunye nootitshala baseyunivesithi. N.I. Lobachevsky, ngoko udliwano-ndlebe luye lwaba luninzi kwaye lungaqhelekanga. Ngokombono wodliwano-ndlebe, kwakunomdla ukufumanisa ulwazi olusisiseko lobugcisa lwabafundi ngokunxulumene nombono wekhompyutha, ngoko ke izihloko ezifana ne-C++/STL, i-OOP, i-algorithms esisiseko kunye nezakhiwo zedatha, i-algebra yomgca, uhlalutyo lwemathematika, imathematika ecacileyo kunye kwabuzwa okuninzi. Phakathi kwemisebenzi, eyona nto yayiphambili yayikukufumanisa indlela abafundi abaqiqa ngayo. Ikhomishini ikwanomdla wokuba bafunde phi, ngawaphi amava ababenawo phambi kwesi sikolo (umzekelo, imisebenzi yezenzululwazi) kunye nokuba ingasetyenziswa njani ngqo kwicandelo lombono wekhompyutha.
Bebonke ngabafundi abangama-78 abathathe inxaxheba kukhetho olusisigxina, lo gama bekukho izithuba ezisisigxina ezingama-24. Ubalo lwabathathi-nxaxheba kunye nomahluko obonakalayo phakathi kweendlela ezisisigxina kunye nezexeshana zentatho-nxaxheba zinokubonwa kwitheyibhile engezantsi:
Benza ntoni abafana kwiiveki ezi-2?
Abafundi baqhelana nethiyori kunye nokuziqhelanisa neyona misebenzi iphambili yombono wekhompyuter: ukuhlelwa kwemifanekiso, ukubonwa kwezinto kunye nokulandelela kwabo. Icandelo lesifundo kwisihloko ngasinye lidla ngokubandakanya uhambo lwembali ekuphuhliseni iindlela zakudala zokusombulula iingxaki zombono wekhompyuter kunye neendlela zanamhlanje zokusombulula kusetyenziswa umatshini wokufunda kunye nothungelwano lwe-neural. Ithiyori yalandelwa kukuziqhelanisa, apho abafundi bakhuphela iimodeli zothungelwano lwe-neural ezithandwayo kwaye bazisungula besebenzisa imodyuli ye-DNN yethala leencwadi le-OpenCV, besenza isicelo esiqhelekileyo.
Iintetho zazo zonke iintetho zaxhonywa kwindawo yokugcina uluntu , ukuze abafundi bahlale bevula baze bajonge ulwazi oluyimfuneko, kuquka ukuphuma kwesikolo. Kwakunokwenzeka ukunxibelelana nabahlohli, ootitshala abaziqhelanisayo kunye neenjineli ze-Intel bobabini bahlala kwaye ngencoko kwiGitter. Ixesha leveki yeprojekthi nayo yaba yimpumelelo: yaqala ngoLwesithathu, eyenza ukuba kube lula ukuchitha impelaveki ngaphandle kweentetho, ukuphucula izigqibo zeqela. Abona bathathi-nxaxheba banoxanduva bachitha isiqingatha soMgqibelo kwiofisi ye-Intel, apho bavuzwa ngohambo olungacwangciswanga kwangolo suku lunye.
Kwakunjani ukukhuselwa kweeprojekthi?
Iqela ngalinye lanikwa imizuzu eli-10 ukuba lithethe ngento eliyenzileyo ngexesha leprojekthi kunye nento elize kuyo. Emva kweli xesha, imizuzu emi-5 yaqala, apho iinjineli zenkampani zabuza abafana imibuzo kwaye zanika iingcebiso ezincinci eziza kubanceda baphucule iprojekthi yabo okanye bathintele iimpazamo ezikhoyo kwixesha elizayo. Ngamnye wabafana bazama ngokwabo njengesithethi, bebonisa ulwazi lwabo kwimbono yekhompyutheni kunye nokuqinisekisa igalelo labo ekudalweni kweprojekthi, eyasinceda ukuba sicinge kwaye senze isigqibo malunga nomthathi-nxaxheba ngamnye esikolweni. Ukhuselo lwenzeka ngaphezulu kweeyure ze-3, kodwa sabakhathalela abafana kwaye sanciphisa uxinzelelo kunye nekhefu elifutshane lekofu, apho abafana banokuphefumla kwaye baxoxe ngemiba kunye neengcali ezikhokelayo ze-Intel.
Ekupheleni kosuku, sawonga indawo yokuqala, yesibini eyesibini, neyesithathu. Kwakunzima kakhulu ukukhetha, kuba iqela ngalinye, iprojekthi nganye inencasa yayo kwaye yahlulwa ngemvelaphi yayo yokubonisa.
Abathathi-nxaxheba be-CV bexesha elizeleyo, ukukhusela iprojekthi, iofisi ye-Intel eNizhny Novgorod
Iiprojekthi ezinikezelweyo
Iglavu ehlakaniphile
Ukusebenzisa i-detector kunye ne-tracker usebenzisa i-OpenCV yokukhangela okubonakalayo kwindawo. Iqela longeze ubunzulu bokuva ubunzulu kusetyenziswa iikhamera ezimbini. IMicrosoft Speech API isetyenziswa njengojongano lolawulo.
Isamkeli
Ukufunyanwa kokutya kunye nokukhethwa kweresiphi yesitya esele silungile, kubandakanywa izithako ezifunyenweyo. Abafana babengoyiki umsebenzi kwaye ngaphakathi kweveki baphawula inani elaneleyo lemifanekiso ngokwabo, baqeqeshe umtshina usebenzisa i-API ye-TensorFlow Object Detection kwaye yongeza ingqiqo yokufumana iresiphi. Ilula kwaye inencasa!
Umhleli 2.0
Abathathi-nxaxheba beprojekthi basebenzise isethi yothungelwano lwe-neural (ukukhangela ubuso, ukuqhelaniswa komfanekiso wobuso ngamanqaku aphambili, ukubalwa kwenkcazo yomfanekiso wobuso) ukuqaphela ubuso njengenxalenye yomsebenzi wokukhangela amaqhekeza kwiividiyo ezinde apho umntu othile ekhona. ekhoyo. Inkqubo ephuhlisiweyo ingasetyenziselwa njengenkqubo yokuncedisa ukulungiswa kwevidiyo, ukukhulula umntu ukuba abukele ividiyo ngokwakhe ekufuneni iziqwenga eziyimfuneko. Ukusebenzisa iinethiwekhi ze-neural ukusuka , iqela likwazile ukufezekisa isantya esiphezulu sesicelo: kwi-laptop ene-Intel Core i5 iprosesa, isantya sokucubungula ividiyo sasiyi-58 izakhelo ngesibini.
Isichazi-magama
Ukuzoba iiglasi kunye nemaski ebusweni bomntu. Inethiwekhi ye-MTCNN yayisetyenziselwa ukukhangela ubuso kunye namanqaku abalulekileyo.
Ongaziwayo
Omnye umsebenzi onomdla kwisihloko sokufihla isazisi. Eli qela lazisa iinketho ezininzi zokuguqula ubuso: ukufiphala kunye ne-pixelation. Kwiveki enye, abafana abazange baqikelele nje umsebenzi, kodwa babonelele indlela yokufihla umntu othile (ngokuqaphela ubuso).
Shushu
Iqela leprojekthi "Ukufudumala" lisombulule ingxaki yokudala umncedisi wezemidlalo kwi-head tilt exercise. Kwaye nokuba isicelo sokugqibela sesi sicelo sisenempikiswano, uphando olubanzi lwenziwa ngokuthelekisa iindlela ezahlukeneyo zokubona ubuso: I-Haar cascades, inethiwekhi ezivela kwi-TensorFlow, i-OpenCV kunye ne-OpenVINO. Safudumala kungekuphela nje emzimbeni, kodwa nangengqondo!
Ngaphantsi kwama-800
I-Nizhny Novgorod, isixeko apho isikolo senzeke khona, siya kuguqula iminyaka engama-2 kwiminyaka emi-800, oku kuthetha ukuba kukho ixesha elaneleyo lokuphumeza iprojekthi enomdla. Sacela abantwana ukuba bacinge ngomsebenzi wokudala isikhokelo, ngokusekelwe kumfanekiso we-facade yezakhiwo, unokubonelela ngolwazi malunga nokuba luhlobo luni lwento eboniswa emfanekisweni kwaye zeziphi iinyani ezaziwa malunga nalo. Ngokombono wethu, lo msebenzi wawuyeyona nto inzima kakhulu, kuba inxulumene nombono wekhompyutheni yeklasi, kodwa iqela libonise umphumo ohloniphekileyo.
I-Rock Paper Scissors
Ngaphandle kwemithintelo yexesha elingqongqo lokugqibezela umsebenzi woyilo, eli qela alizange loyike ukwenza umfuniselo ukuze liqeqeshe eyalo inethiwekhi ye-neural ukuhlela izikhundla zezandla kumdlalo owaziwayo.
Ingxelo evela kubathathi-nxaxheba
Sicele abafundi abavela kwizifundo ezahlukeneyo ukuba babelane ngezimvo zabo ngeSikolo saseHlobo:
Kutshanje ndibe nethamsanqa lokuzimasa i-Intel Computer Vision Summer Camp kwaye ibingamava amnandi. Siye safumana ulwazi olutsha kunye nezakhono kwinkalo yeCV, ukufakwa kwesoftware, ukulungisa iimpazamo, siye santywiliselwa kwindawo yokusebenza, sajongana neengxaki zokwenene, saxoxa ngezisombululo ezinokwenzeka kunye noogxa kunye nootitshala besikolo ibandakanya kuphela ukunxibelelana nekhompyuter. Nangona kunjalo, oku akunjalo konke konke. Umsebenzi wethu wokudala awunakwahlulwa kunxibelelwano nabantu. Kwakungonxibelelwano awathi ubani afumane ulwazi olukhethekileyo. Kwaye bendilithanda kakhulu eli candelo lesikolo. Nangona kunjalo, kukho enye i-drawback ... emva kokugqiba izifundo zam ndifuna ukuqhubeka! Ukongeza kulwazi lwethiyori kwi-DL kunye nezakhono ezisebenzayo kwi-CV, ndiye ndafumana umbono wokuba zeziphi iindawo zemathematika ekufuneka zinikwe ingqwalasela eyodwa kwaye zeziphi itekhnoloji ekufuneka ifundwe. Ukuzinikela, ubugcisa kunye nothando ngomsebenzi wabo weenjineli ze-Intel kunye nabaphandi baphembelela ukhetho lwam lolwalathiso kwi-IT. Kungenxa yoku ndithanda ukubulela bonke abaququzeleli besikolo.
UKristina, unyaka wokuqala, HSE
Ngexesha elifutshane kangako, isikolo sibe nako ukubonelela ngolwazi oluphezulu kunye nokuziqhelanisa nesihloko sombono wekhompyutha. Kwaye nangona yayiyilelwe ulwazi olusisiseko, iintetho zaziqulathe izinto ezininzi zobugcisa ofuna ukuziqonda kwaye uchithe ixesha elininzi ufunda. Abacebisi besikolo nabahlohli baphendula ngomdla yonke imibuzo kwaye banxibelelana nabafundi. Ewe, ngelixa ndigqibezela iprojekthi yokugqibela, kuye kwafuneka ndingene kwihlathi ukuze ndiphuhlise isicelo esigqityiweyo kwaye ndihlangabezane nobunzima obungasoloko buvela xa ndifunda. Iqela lethu ekugqibeleni lenze isicelo sokudlala umdlalo "i-rock-paper-scissors" ngekhompyuter. Siqeqeshe imodeli ukuqaphela umzobo kwikhamera yewebhu, sabhala ingqiqo kwaye senza ujongano olusekwe kwisikhokelo se-opencv. Isikolo sabonelela ngokutya kokucinga kunye nevector yokufunda nophuhliso olulandelayo. Ndiyavuya kuba ndithathe inxaxheba.
USergey, unyaka wesi-3, UNN
Isikolo asizange sifikelele kanye njengoko ndandilindele. Iintetho zanikwa ngabantu abanamava ngokufanelekileyo abaphuhlisi be-Intel. Unxibelelwano nabahlohli belusoloko lunomdla kwaye luluncedo, abacebisi bayasabela kwaye bahlala belungele ukunceda. Kodwa sele ndizazi ezinye izinto, kwaye ezo endingaziyo azizange zixhaswe ngokuziqhelanisa nangayiphi na indlela, kwaye ngoko ke izinto ezilungileyo ngokwenene azizange ziqondwe ngokupheleleyo kwaye zifundwe ndim. Ewe, uninzi lolwazi lubonelelwe ngeenjongo zolwazi, ukuze emva koko uzame ekhaya, okanye ube nombono wento yonke malunga nayo, kodwa bendisafuna ukuphumeza ii-algorithms ezikhoyo ngokwam phantsi kwe ukubeka iliso kootitshala abanamava abanokunika iingcebiso ezifanelekileyo okanye uncedo xa kukho into engenzekiyo. Ngenxa yoko, ekusebenzeni, izisombululo esele zenziwe zisetyenzisiwe, kwaye ikhowudi, umntu unokuthi, yayibhalwe ngaphambili kuthi; Iiprojekthi zaziyezona zilula, kwaye ukuba uzama ukuntsonkotha umsebenzi ngandlela thile, ngoko awunalo ixesha elaneleyo lokuwuphumeza ukuya kwisimo esizinzile ngakumbi okanye esingaphantsi, njengoko kwenzeka kuthi.
Ngokubanzi, isikolo siphela sijongeka ngathi luhlobo oluthile lomdlalo ongekho mbi kakhulu wabaphuhlisi, kwaye oku kuyimpazamo yecandelo elisebenzayo. Ndicinga ukuba kuyimfuneko yokwandisa ixesha elichithwe esikolweni, ukudibanisa izinto zokuziqhelanisa ukuze ukwazi kwaye kufuneka ubhale into ngokwakho, into enzima ngokwenene kwaye iyimfuneko, kwaye ungasebenzisi esele zenziwe, ukwenza umkhuba ube lula ekwandeni. ubunzima, izihloko zeeprojekthi zokhuphiswano kufuneka zinikezelwe ngeentsuku zokuqala, ukwenzela ukuba izinto ezivela kwiintetho kunye nezenzo zisetyenziswe ngokukhawuleza kwiiprojekthi zakho kwaye kuya kuba nexesha elingakumbi lokuphunyezwa. Emva koko ixesha elichithwe esikolweni liya kusebenza njengamava alungileyo kwabo bafuna iingcali.
UDmitry, unyaka wokuqala wesidanga se-masters, i-NSTU
Isikolo sasehlotyeni esivela kwi-Intel yayilithuba elihle lokuchitha eli hlobo usenza into oyithandayo. Inyaniso yokuba iintetho zanikwa ngabasebenzi be-Intel ezinxulumene neprogram kwintsimi yombono wekhompyutheni ayizange indivumele ukuba ndiphumle; Yonke imihla yayidlula ngokukhawuleza, ingabonakali kwaye ineziqhamo. Ithuba lokuphumeza iprojekthi yam lindivumele ukuba ndisebenze kwiqela elinabagcini abamangalisayo kunye nabanye abathathi-nxaxheba besikolo. Ezi veki zimbini zinokuchazwa ngokufutshane ngolu hlobo lulandelayo: umdla kwaye uyakhawuleza.
U-Elizaveta, unyaka we-2, UNN
Ekwindla (ngo-Oktobha-Novemba), inkqubo yemfundo ye-Delta ilindele wena, ulwazi malunga nokuba ungalufumana kwithu . Hla umamele!
umthombo: www.habr.com