I-LDM3D, imodeli yokuqala yokusabalalisa yemboni ukunikeza ukujula kwemephu ukuze udale izithombe ze-3D ezinokubukwa okungu-360-degree okucacile futhi okugxilile.
I-Intel ne-Blockade Labs ikhululiwe ngolwazi lokuthunyelwe kwebhulogi mayelana nokuthuthukiswa kwabo ngokuhlanganyela kwemodeli yokufunda yomshini ebizwa ngokuthi "I-LDM3D» (Imodeli Ecashile Yokusabalalisa ye-3D) ukukhiqiza izithombe nokujula kwamamephu izixhumanisi ezisekelwe ezincazelweni zombhalo wolimi lwemvelo.
Imodeli yaqeqeshwa kusetshenziswa isethi yedatha evuliwe ye-LAION-400M. Ilungiselelwe umphakathi we-LAION (Large-Scale Artificial Intelligence Open Network), othuthukisa amathuluzi, amamodeli, nokuqoqwa kwedatha ukuze kwakhiwe izinhlelo zokufunda zamahhala zemishini. Iqoqo le-LAION-400M lihlanganisa izithombe eziyizigidi ezingu-400 ezinezincazelo zombhalo.
Ngaphezu kwezithombe nezincazelo zazo zombhalo, amamephu ajulile nawo asetshenziswa lapho kuqeqeshwa imodeli ye-LDM3D, ekhiqizwe isithombe ngasinye kusetshenziswa uhlelo lokufunda lomshini lwe-DPT (Dense Prediction Transformer), okuyinto ikuvumela ukuthi ubikezele ukujula okuhlobene kwephikseli ngayinye wesithombe esiyisicaba.
I-Intel Labs, ngokubambisana nama-Blockade Labs, yethule i-Latent Diffusion Model ye-3D (LDM3D), imodeli yokuqala yokusabalalisa yemboni enikeza ukujula kwemephu yokwakha izithombe ze-3D ezinokubukwa kwe-360-degree okucacile futhi okugxilile. .
I-LDM3D inamandla okuguqula ukudalwa kokuqukethwe, izinhlelo zokusebenza ze-metaverse, nolwazi lwedijithali, iguqule inhlobonhlobo yezimboni, kusukela kwezokuzijabulisa nemidlalo kuya ekwakhiweni kwezakhiwo nokuklama.
Uma kuqhathaniswa nobuchwepheshe bokubikezela obujulile ekucutshungulweni kwangemuva, imodeli I-LDM3D, ekuqaleni waqeqeshwa ngokujulile, inikeza ulwazi olunembe kakhudlwana esigabeni sesizukulwane. Enye inzuzo yemodeli yikhono lokukhiqiza idatha yokujula ngaphandle kokwandisa inani lamapharamitha: inani lamapharamitha kumodeli ye-LDM3D cishe lifana nemodeli yakamuva yokusabalalisa ezinzile.
Ukukhombisa amakhono yemodeli Isicelo se-DepthFusion sesilungisiwe, ukuthi ikuvumela ukuthi udale izindawo ezisebenzisanayo zokubukwa ngemodi ye-360 degree kusukela kuzithombe ze-RGB ezinezinhlangothi ezimbili namamephu ajulile.
I-LDM3D ivumela abasebenzisi ukuthi benze isithombe kanye nemephu ejulile kusukela kumlayezo wombhalo onikeziwe usebenzisa cishe inombolo efanayo yamapharamitha.
I-LDM3D ibhalwe ku-TouchDesigner, ulimi lokuhlela olubonakalayo olufanele ukudala okuqukethwe kwe-multimedia okusebenzisanayo ngesikhathi sangempela. Imodeli ye-LDM3D ingase futhi isetshenziselwe ukukhiqiza nokuguqula izithombe ngokusekelwe kusifanekiso esihlongozwayo, iphrojekthi umphumela endaweni ukuze kwakhiwe indawo ezungezile, ukukhiqiza izithombe ngokusekelwe ezindaweni ezihlukene zezibukeli, futhi kukhiqizwe ividiyo esekelwe ekunyakazeni kwekhamera ebonakalayo.
Ubuchwepheshe obuhlongozwayo kufanele bube namandla amakhulu okudala izindlela ezintsha yokusebenzisana kwabasebenzisi, okungaba yimfuneko ezimbonini ezihlukahlukene, kusukela kwezokuzijabulisa nemidlalo kuya ekwakhiweni kwezakhiwo nokuklama. Isibonelo, i-LDM3D ingasetshenziselwa ukudala amamnyuziyamu asebenzisanayo kanye nezindawo ezingokoqobo ezingokoqobo ezikhiqiza indawo enemininingwane esekelwe ezifisweni zolimi lwemvelo.
Intuthuko ifana nesistimu ye-Stable Diffusion synthesis yesithombe, kodwa ivumela ukwakheka kokuqukethwe okubukwayo kwezinhlangothi ezintathu, njengezithombe zepanoramic eziyindilinga ezingabukwa ngemodi engu-360-degree. Ngasohlangothini olusebenzayo, imodeli ingasetshenziswa kumageyimu namasistimu wento engekho ngokoqobo ngokwakhiwa okusebenzisanayo kwezindawo ezinezinhlangothi ezintathu.
Imodeli ye-LDM3D iqeqeshelwa i-Intel AI supercomputer enama-Intel® Xeon® processors nama-accelerator e-Intel® Habana Gaudi® AI.
Kulabo abathanda iphrojekthi, kufanele bakwazi lokho imodeli elungele ukusetshenziswa inikezwa ukuze ilandwe mahhala kumasistimu okufunda emishini, okuthi ingasetshenziswa nge-PyTorch kanye nekhodi eklanyelwe ukukhiqiza izithombe kusetshenziswa amamodeli avela kuphrojekthi ye-Stable Diffusion.
okufanele kukhulunywe ngakho ukwedlula imodeli isatshalaliswa ngaphansi kwelayisensi yemvume Creative ML OpenRAIL-M, okuyinto ivumela ukusetshenziswa kwezohwebo. Ukusabalalisa ngaphansi kwelayisense evulekile kuvumela abacwaningi abanentshisekelo nonjiniyela ukuthi bathuthukise imodeli ngokuya ngezidingo zabo futhi bayisebenzisele izinhlelo zokusebenza ezikhethekile.
Ekugcineni, uma unentshisekelo yokwazi kabanzi ngakho, ungaxhumana nemininingwane Kulesi sixhumanisi esilandelayo.