Iqembu labacwaningi abavela emanyuvesi amaningana aseMelika lishicilele iphrojekthi (I-Pixel-Aligned Implicit Function), ekuvumela ukuthi usebenzise izindlela zokufunda zomshini ukuze wakhe imodeli ye-3D yomuntu kusukela kusithombe esisodwa noma ngaphezulu esinezinhlangothi ezimbili. Uhlelo lukuvumela ukuthi udale kabusha izinketho zokugqoka eziyinkimbinkimbi, njengeziketi ezicwebezelayo nezithende, nezitayela zezinwele ezihlukahlukene, ngokuzimela ukubuyisela ukuthungwa nokuma ezindaweni ezingabonakali ekuboneni lapho imodeli ye-3D yakhiwe khona. Ukwandisa ikhwalithi kanye nemininingwane yemodeli yokugcina ye-3D, izithombe ezimbalwa ezivela ema-engeli ahlukene zingasetshenziswa. Ikhodi yephrojekthi ibhalwe kuPython kusetshenziswa uhlaka lwePyTorch kanye ngaphansi kwelayisensi ye-MIT.
Inethiwekhi ye-neural isetshenziswa njengomthombo wokwakha kabusha isakhiwo esinezinhlangothi ezintathu, esikuvumela ukuthi ukhethe umumo okungenzeka kakhulu futhi usungule izakhi ezifihliwe, kusukela kumodeli eqeqeshwe ngezinguqulo ezihlukahlukene zezinto ezikhona. Ngokuhambisanayo, iphrojekthi ihlinzeka nge-algorithm yokufanisa isakhiwo se-volumetric esiwumphumela nemibala ezithombeni ze-2D ezinikeziwe, eziqondanisa amaphikseli esithombe se-3D ngokuya ngokuma kwawo entweni ye-XNUMXD futhi ikhiqize ukuthungwa okungenzeka ukuthi akukho. Noma yisiphi isithombe singabhalwa ngekhodi ngoba
i-surface reconstruction esetshenziswa izakhiwo "", a
Inethiwekhi ye-neural esekelwe kuzakhiwo isetshenziselwa ukufanisa ukuthungwa .
Imodeli eyenziwe ngomumo eqeqeshiwe esetshenziswa abacwaningi iyatholakala ukuze ilandwe mahhala, kodwa idatha eluhlaza esetshenziselwa ukuqeqeshwa ihlala iyimfihlo njengoba isekelwe kumaskeni we-3D okuhweba. Ingasetshenziswa njengomthombo wokuziqeqeshela imodeli abantu abavela kuphrojekthi ye-Renderpeople.
Source: opennet.ru