Uma ungabonanga ukushisa kwe-cranium, okunikeza ucwaningo lukaKevin Karsch, bayibuke, sizolinda... Manje njengoba ubuchopho bakho sebushile futhi buwohloka kuhle kwamaphophukhoni ashile, ungaqonda ukuthi kungani, ngemva kokubona le vidiyo, bekufanele thola okuphansi kuKevin uqobo. Ngakho... samgoqa angaboni, samphenya, samfaka izidakamizwa, sampakisha esitsheni semikhumbi, samthumela kuSolidSmack HQ ukuze ayophenywa (akusilo iqiniso). Nakuba ukumodela okusekelwe ezithombeni kuyinto engaziwa, u-Kevin nethimba basondela ezindleleni ezintsha zokufaka izinto ze-3D ezigcawini ze-2D, futhi njengoba u-Kevin echaza, izithombe ziyisiqalo.
Ukudala Isigcawu
Kevin Karsch njengamanje ungumfundi we-PhD eNyuvesi yase-Illinois futhi uzokwethula ucwaningo ngalo Disemba ngo SIGGRAPH ASIA 2011. Njengoba i-abstract isho, lokhu kunjalo "indlela yokufaka izinto zokwenziwa ezithombeni ezikhona ngaphandle kokudinga ukufinyelela endaweni yesehlakalo noma ezinye izilinganiso zesigcawu ezengeziwe." Nakuba abanye bengakhetha lokhu ekuhleleni izithombe okulula, ubuchwepheshe kanye namathuba adlulela ngale kwalokho.
I-SolidSmack: Iyini isizinda sakho, futhi uqale kanjani ukubhala isofthiwe yehluzo?
U-Kevin Karsch: Ngicabanga ukuthi ngibhale uhlelo lwami lokuqala lwehluzo esikoleni esiphakeme; bekuyindlela elula enamahloni ye-Pong :). Lesi sifundo siholele ekutheni ngiphishekele isayensi yekhompiyutha kanye nokuthuthukiswa komdlalo wevidiyo ukuze ngithole iziqu zami zeziqu eNyuvesi yaseMissouri, nokuyilapho ngaqala khona ukwenza ucwaningo ngemifanekiso kanye nombono wekhompyutha (okungukuthi ukukhipha ulwazi olusezingeni eliphezulu ezithombeni kanye/noma kumavidiyo) ucwaningo. Manje, ngiyikhandidethi ye-PhD eNyuvesi yase-Illinois ngisebenza noProf. David Forsyth kanye noProf. Derek Hoiem, futhi ucwaningo lwami engigxile kulo luyingxube yakho kokubili imidwebo nombono.
SS: Yikuphi okukuthakasele kuqala ekunikezeni into ngokuqondile ezithombeni eziyi-2d?
I-KK: Umqondo wokukwazi ukusebenzisana nesithombe ngendlela efanayo noxhumana ngayo nesimo somzimba wawusijabulisa kakhulu. Izicelo eziningi ziyavela uma lokhu kungenzeka; hhayi nje ukufaka izinto, kodwa ukususa izinto ezikhona, ukuguqula izakhiwo zezinto ezibonakalayo, ukulungisa izibani, njalonjalo. Sinqume ukuhlasela inkinga yokufaka kuqala.
Abantu bahle kakhulu ekunqumeni isimo sokwakheka komzimba ngokubuka nje isithombe, kodwa ukwenza lokhu kuhlelwa kungaba nzima kakhulu futhi kukhathaze ngamathuluzi amanje. Umgomo wethu bekuwukuvumela abasebenzisi ukuthi benze kanjalo ngokushesha nangendlela enembile, esikwazile ukuyifinyelela ngokuhlanganisa ndawonye ama-algorithms amaningi ocwaningo asezingeni eliphezulu (okunye okwethu, kanti amanye akhona).
SS: Ungakwazi yini ukuchaza ubuchwepheshe obusemuva kwenqubo ngokwemibandela yabantu abangajwayelekile?
I-KK: Ukuze sifake imodeli ye-3D esithombeni, sidinga ukumelwa kwe-3D kwesigcawu, okuhlanganisa i-geometry, izibani, izici ezibonakalayo, namapharamitha ekhamera (ubude bokugxila, indawo yekhamera, njll). Umsebenzisi unikeza ulwazi oluthile lwezinga eliphezulu, njengemingcele yesigcawu kanye nendawo yemithombo yokukhanya ngokumaka esithombeni. Ngalolu lwazi, singakwazi ukubala ngokuzenzakalelayo imodeli engalungile yesehlakalo. Amapharamitha wejiyomethri nekhamera ahlanganiswa kusetshenziswa indlela yokuthola isakhiwo se-3D kumaphoyinti angu-2D esithombeni; lokhu kuvame ukubizwa ngokuthi yi-single view metrology (bona i- Iphepha le-IJCV nguCriminsi et al.). Sisebenzisa i-geometry nekhamera, sikhetha (ngokuthuthukisa izinombolo) amapharamitha wezinto ezingcono kakhulu kanye nezindawo zomthombo wokukhanya ukuze isithombe esinikeziwe sesigcawu sethu esakhiwe kabusha se-3D sifane kangcono nesithombe sokuqala. Kukhona futhi imininingwane ethile yokubhekana namashafti okukhanya nemingcele yento evalekile, futhi siyakwazi ukulinganisa amamodeli alawa ngemva kwenani elincane lemakhaphu yomsebenzisi.
SS: A iphepha lokumodela okusekelwe ezithombeni futhi ukuhlelwa kwesithombe kwashicilelwa ngo-2001 lokho futhi kusebenzisa umbono wesithombe esisodwa njengokufakwayo. Luhluke kanjani ucwaningo lwakho futhi uhlelo lokusebenza luzosetshenziswa nini ngqo kwisoftware ehlinzeka nge-3D?
I-KK: Iphepha elishiwo linikeza isethi enkulu yamathuluzi okuthola amamodeli anembe uma kuqhathaniswa ejiyomethri ukusuka kwesinye isichasiselo somsebenzisi, futhi libonisa indlela yokunquma izinto ezibonakalayo ezifana nokubonisa kusetshenziswa ijometri. Iphepha lethu empeleni lisebenzisa indlela efanayo kakhulu ukulinganisa izakhiwo zezinto ezibonakalayo, kodwa sidinga imodeli enemininingwane encane kakhulu yejometri, futhi ngaleyo ndlela sidinga isichasiselo esincane kakhulu. Umehluko oyinhloko ukuthi siphinde silinganisele amapharamitha ekhamera nolwazi lokukhanyisa, okuwukhiye ekufakeni izinto zokwenziwa. Sithole ukuthi amamodeli alula ejiyomethri anele ukufakwa okuningi, ngoba amaphutha avela kuphela lapho izinto ezifakiwe zihlangana nejometri eyenziwe ngendlela engalungile (futhi kuye kwashiwo ezincwadini ze-psychophysical ukuthi abantu abalungile kakhulu ekuthatheni lezi zinto. ukungahambisani). Kodwa-ke, iqoqo lamathuluzi ethulwe kuleli phepha lizohambisana ngendlela emangalisayo nendlela yethu uma umuntu edinga i-geometry enembe kakhudlwana, mhlawumbe ukulingisa okungokoqobo.
Singathanda ukubona lobu buchwepheshe bufakwa kumodeli ye-3D nesofthiwe yokunikezela ngokushesha okukhulu. Sisebenzisana nenyuvesi yethu ukwenza lokhu kwenzeke.
I-SS: Empini yokunikezwa kwe-CPU vs GPU, ama-algorithms okukhanya omhlaba ayinkimbinkimbi aba yinkimbinkimbi nakakhulu uma wengeza isilinganiso sokukhanya nezinto ezibonakalayo. Njengoba ubuchwepheshe bokunikezela buqhubeka, ingabe izingxenyekazi zekhompuyutha zizothuthuka kakhulu ekucubunguleni okunemicu eminingi noma iphrosesa yezithombe?
I-KK: Kimina, ngokombono wocwaningo, lokhu kuncike ngempela ekutheni ijika lokufunda liwumqansa kangakanani wokuhambisana kuma-CPU nama-GPU azayo. Njengamanje, kubonakala sengathi ama-CPU ayayinqoba le mpi, futhi ngikholwa ukuthi yingakho iningi lesoftware ehlinzeka namuhla ibhalelwe ama-CPU. Uma kubhekwa lo mkhuba, kubonakala sengathi ukucubungula okunezintambo eziningi kuzobusa eminyakeni embalwa ezayo. Kodwa-ke, lokhu kungashintsha uma ikhodi yokubhala nokulungisa iphutha kuma-GPU kungenziwa ngendlela elula efana neyakuma-CPU. Isixazululo sesithathu esingase sivele i-hardware ekhethekile (ingxube ethile yalokhu ekhona namuhla) eyakhelwe ngqo ukunikezela.
SS: Isinyathelo esilandelayo esisobala esokuba ijiyomethri engu-2d ikhishwe ngokuzenzakalelayo futhi inikezwe endaweni elandelwa umsebe wesikhathi sangempela. Uyakubona lokhu okwenzekayo futhi yiziphi izinselelo okufanele zixazululwe ukuze kwenziwe lokhu?
KK: Ngiyavuma; lokhu kungaba nemiphumela emihle yeqiniso elingathandwa kwabathelisi esikubona kanye nenani lezinye izinhlelo zokusebenza! Ngicabanga ukuthi kuzoba ngokoqobo eminyakeni embalwa phansi komgwaqo, kepha ubuchwepheshe obuningi budinga ukuthuthuka futhi buhlangane. Manje ngicabanga ukuthi amakhamera ajulile afinyeleleka kakhulu (isb. I-Microsoft's Kinect), kuyindaba nje yesikhathi ngaphambi kokuthi kube nokwenzeka ukusho ngokuzenzakalelayo imithombo yokukhanya nezinto zokwakha, bese ukujula/ijometri iza mahhala ku-Kinect. Inselele enkulu ingaba ekwenzeni isistimu ibe yisikhathi sangempela, okungahle kudinge ukuthuthukiswa kokusebenza kahle ezigabeni zokulinganisa nezokuhlinzeka, kanye nezingxenyekazi zekhompuyutha ezisheshayo futhi.
SS: Njengoba uthuthukisa noma yibuphi ubuchwepheshe obuyisisekelo njengalobu, nakanjani uzothola izinto obungazilindele ngesikhathi uqhubeka. Kukhona ezinye izimo zokusebenzisa ezichazwe ngokucacile zalobu buchwepheshe njengoba zivezwe kuvidiyo yakho, kodwa ingabe zikhona izinhlelo zokusebenza ezingabonakali ohlangane nazo endleleni? Noma yimiphi imiphumela obungayilindele, kodwa ibe usizo?
I-KK: Uhlelo lokusebenza olulodwa olujabulisayo esizwile ngalo ukusebenzisa lobu buchwepheshe ukufaka izinto ezithombeni zomlando ngezinjongo zokufundisa. Siphinde sathola ukuthi kungase kube usizo ekudaleni ingemuva lezikhangiso ngokushesha, futhi kube namandla okuvumela amafilimu anebhajethi ephansi ukuthi aqhudelane nezinkampani ezinemiphumela ephezulu. Siphinde sibe nemibuzo eminingi mayelana nezindlela zethu zamavidiyo (kunokuba abe nezithombe ezimile), okubonakala sengathi unezinhlelo zokusebenza ku-real estate, idizayini yezakhiwo, nokuhlobisa kabusha ikhaya, phakathi kokunye.
Njengamanje sihlanganisa ividiyo yobufakazi bomqondo ebonisa isandiso esilula endleleni yethu evumela ukuthi izinto zifakwe ngokunembile kumavidiyo angadingi okokufaka okwengeziwe okuvela kumsebenzisi, ukuze sikwazi ukubona lezi zinhlelo zokusebenza ngokushesha saqaphela. Sinenqwaba yemibono yocwaningo lwangomuso, futhi ngethemba ukuthi lo msebenzi uzokhuthaza amanye amaqembu ukuthi ahlole imibono emisha ngalesi sihloko!
-
Sibonga kakhulu kuKevin ngokuxoxa nathi ngocwaningo lwakhe. Uma ungathanda ukuthola okwengeziwe, ungavakashela ku- ikhasi lephrojekthi nge-abstract kanye nokushicilelwa okungena emininingwaneni eminingi mayelana nokufaka izinto ezigcawini kanye nokuthi kwenziwa kanjani.
