Zero-1-ba-3 Tasaaroo gis-gis bu bees
Zero-1-to-3 dafay soppi benn nataalu benn mbir mu nekk nataalu benn mbir buñu xoolee ci angle bu bees, jëfandikoo modelu diffusion bu lalu ci wëréelu kamera bi nga laaj.
Résumé
It matters because it lets you reconstruct 3D-consistent views without ever scanning the object from multiple sides.
Plongeur bu xóot
Zero-1-ba-3 (ci Columbia, 2023) dafay gëna rafetal diffusion bu dëgër suko defee mu mëna def synthese bu bees bu amul benn nataal ci benn nataal buñ dugal. Danga koy jox benn nataal boole ci soppi kamera bi (rotaasioŋ ak tekki bu ndaw), model bi dafay defar ni mbir mi di nuru ci gis-gis bu bees boobu. Li gëna am solo mooy xeetu diffusion 2D yu mag yi, ñu tàggat ci dajale nataali web yu mag, dañu jël bu baax geometrik ak physique ci ni mbir yi di nuru ci 3D. Suñu defaree ay done yuñ defaree ay mbir yuñ defaree ci angle kamera yu bari yuñ kontre (jëfandikoo Objaverse), model bi dafay jàng ni ñuy xaymaa prior yooyu ci kontre kamera bu leer. gis-gis yiñ defar mën nañu dundal tabaxaat 3D bi ci suuf.
Gis-gis xarala
Modèle bi dafa am ñaari anam ci nataalu source bi: ab CLIP embedding dafay boole ak pose kamera bi (azimuth, elevation, rayon) ngir dindi cross-attention, ci noonu la nataal bu ñor bi di chaine-concatené ci latent bu bari bruit bi suko defee ay detay yu ndaw ak dàntite ñu ngi denc. Taggat dafay jëfandikoo ñatti nataal-pose-nataal yuñ defaree ci mbir CAD, suko defee reso bi jàng kàrt biñ mëna saytu diggante coppite ci wàllu xool ak coppite ci pixel yi ciy juddoo.
njeextalu pexe
Gaawaay ak yaatuwaay
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal.
Tabax tànneef
Ekipu kreatif yi mën nañu defar konsept yu gëna gaaw te duñu def lu bari ci loxo.
Ekip ak def liggéey
Liggéeyukaay yi mën nañu jëfandikoo siñaal nataal wala wideo yu jafewoon lool ci liggéey.
Ëlëgu Zero-1-ba-3 Tasaaroo gis-gis bu bees
Zero-1-ba-3 moo jiite vague bu nataal-ba-3D. Ñi ko wuutu lu melni Zero123-XL, SyncDreamer, ak One-2-3-45 dañuy puus ci wàllu gis-gis yu bari ak gëna gaaw, gëna wóor 3D mesh, ci noonu la boole ak Gaussian Splatting ak xeetu tabaxaat yu mag di wàññi diiru defar ci ay simili ba ay segond. Xaarandil gis-gis bu gëna dëgër, dayo bu gëna kawe, ak àdduna dëgg (bañ rek objet synthetik) generalisation ndax yii xeetu diffusion yuñ mëna saytu ci wàllu gis-gis dañuy màgg nekk jumtukaay yuñ miin ngir defar ëmbiit.
Doxal ci àdduna dëgg
Defar ay nataalu benn produit ci plateforme bi suko defee limu e-commerce bi mëna wane produit bi ci wàll yépp
Bootstrapping ab mesh 3D bu am tekstuur bu ab mbir ci benn nataalu telefon bu yomb ngir ay preview AR
Sosal art bu méngoo ak wàll yu bari ci ab personage wala prop ngir jeu video ak artist konsept film
Dundal gis-gis yu bees yiñ defar ci NeRF wala tabaxaat Gaussian Splatting ngir feesal geometrie buñu gisul
Risk yi ak balustrade yi
Yelleefi nataal ak nangu mën na nekk risku yoon sudee fi ñu bawoo leerul.
Performance model bi mën na wuute ci leeraay bi, demographie bi ak environmaa bi.
Njuumteg positive yi mën nañu dem te kenn duko seetlu fileek xool wuñu buntu wóolu sa bopp.
Roadmap ngir samp gi
Mandargal kritërium nangug njub, woowaat ak njëgu njuumte.
Saytu ak done yu méngoo ak anam yi ñuy liggéeyee dëgg.
Yokk jàngat nit ngir xam fu wóorul dara wala am njeexital yu rëy.
Toppal model drift bi nga baaxal ko ginaaw bi kamera bi wala done yi soppeekoo.
Weyal di banneexu
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Zero-1-to-3 Novel View Diffusion quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Xeetu gis-gis bu bees
Laaj yi ñuy faral di laaj
What is Zero-1-to-3 Novel View Diffusion?
Zero-1-to-3 dafay soppi benn nataalu benn mbir mu nekk nataalu benn mbir buñu xoolee ci angle bu bees, jëfandikoo modelu diffusion bu lalu ci wëréelu kamera bi nga laaj. Dafa am solo ndax daf lay may nga defaraat gis-gis yu méngoo ci 3D te doo musa skane mbir mi ci wàll yu bari.
Lan mooy core input Zero-1-to-3 soxla ci kaw benn nataal ngir sosal gis-gis bu bees?
Zero-1-to-3 dafay aju ci benn nataal boole ci pose kameraa bi, kon nga leeral wëréelu ak tekki bi ci poñ bi nga bëgg.
Zero-1-ba-3 ñu ngi ko tabax ci ajustement bu baax ban xeetu model?
Dafay defar xeetu diffusion 2D bu mag buñu tàggat bu njëkk suko defee mu mëna jëfandikoo priors geometrique yi model yooyu jàngee ci nataali web yi.
Lan moo waral modelu diffusion 2D mëna def synthese bu bees bu amul benn shot?
Taggat 2D bu yaatu dafay joxe xam-xam bu leer ci ni mbir yi di feeñee ci 3D, te fine-tuning dafay tax ñu mëna doxal ci pose kamera.
Ban setu done ci mbir 3D moo nekkoon lu am solo ci tàggat Zero-1-to-3?
Ñu ngi ko tàggat ci ñatti nataal-pose-nataal yuñ defaree ci ay mbir yu bari yuñ dajale ci 3D yu melni Objaverse.
naka lañuy denc detay yu rafet yi ci nataalu cosaan bi ci jamono jii?
Nataal bu ñor bi dafay boole chaine bi ak latent bu bari bruit bi (ak CLIP biñ dugal ci semantik bi) suko defee dàntite bi ak detay yi di jaar.