LDM3D, samfurin yaduwa na farko na masana'antar don bayar da taswira mai zurfi don ƙirƙirar hotunan 3D tare da ra'ayoyi 360-digiri waɗanda suke da haske da nutsewa.
Intel da Blockade Labs sun fito ta hanyar bayanan bulogi game da haɓaka haɗin gwiwa na samfurin koyon injin mai suna "LDM3D» (Tsarin Yaduwa na Latent don 3D) don samar da hotuna da taswira mai zurfi abokan haɗin gwiwa bisa ga bayanin rubutun harshe na halitta.
An horar da samfurin ta amfani da saitin bayanan buɗaɗɗen LAION-400M. LAION (Large-Scale Artificial Intelligence Open Network) al'umma ta shirya, wanda ke haɓaka kayan aiki, samfura, da tarin bayanai don gina tsarin koyo na inji kyauta. Tarin LAION-400M ya ƙunshi hotuna miliyan 400 tare da bayanin rubutu.
Baya ga hotuna da bayanin rubutunsu. Hakanan ana amfani da taswirorin zurfin lokacin horar da ƙirar LDM3D, wanda aka samar don kowane hoto ta amfani da tsarin koyo na injin DPT (Dense Prediction Transformer), wanda yana ba ku damar tsinkayar zurfin dangi na kowane pixel na lebur hoto.
Intel Labs, tare da haɗin gwiwar Labs na Blockade, sun gabatar da Model Diffusion na 3D (LDM3D), ƙirar masana'anta ta farko wacce ke ba da taswira mai zurfi don ƙirƙirar hotuna 3D tare da ra'ayoyi 360-digiri waɗanda suke a sarari da nitsewa. .
LDM3D yana da yuwuwar kawo sauyi ga ƙirƙirar abun ciki, ƙayyadaddun aikace-aikace, da gogewar dijital, yana canza masana'antu da yawa, daga nishaɗi da wasa zuwa gine-gine da ƙira.
Idan aka kwatanta da zurfin tsinkaya fasahar a bayan-aiki, samfurin LDM3D, da farko an horar da su sosai, yana ba da ƙarin cikakkun bayanai masu zurfi a matakin tsara. Wani fa'idar ƙirar ita ce ikon samar da bayanai mai zurfi ba tare da ƙara adadin sigogi ba: adadin sigogi a cikin ƙirar LDM3D kusan iri ɗaya ne da na sabon tsarin watsawar barga.
Don nuna iyawa na samfurin An shirya aikace-aikacen DepthFusion, cewa yana ba ku damar ƙirƙirar yanayin hulɗa don kallo a cikin yanayin digiri 360 daga hotuna RGB masu girma biyu da taswirorin zurfin.
LDM3D yana ba masu amfani damar ƙirƙirar hoto da taswira mai zurfi daga saƙon rubutu da aka bayar ta amfani da kusan adadin sigogi iri ɗaya.
An rubuta LDM3D a cikin TouchDesigner, harshen shirye-shirye na gani wanda ya dace don ƙirƙirar abun ciki na multimedia mai ma'amala a cikin ainihin lokaci. Hakanan za'a iya amfani da ƙirar LDM3D don ƙirƙira da gyaggyara hotuna dangane da samfurin da aka tsara, aiwatar da sakamakon akan wani yanki don ƙirƙirar yanayi, samar da hotuna dangane da matsayi daban-daban na masu kallo, da kuma samar da bidiyo dangane da motsin kyamarar kama-da-wane.
Ya kamata fasahar da ake samarwa tana da babban damar ƙirƙirar sabbin hanyoyin na hulɗar mai amfani, wanda zai iya kasancewa cikin buƙata a masana'antu daban-daban, daga nishaɗi da wasan kwaikwayo zuwa gine-gine da ƙira. Misali, ana iya amfani da LDM3D don ƙirƙirar gidajen tarihi masu mu'amala da mahalli na gaskiya waɗanda ke haifar da cikakkun mahalli dangane da sha'awar harshe na halitta.
Ci gaban yayi kama da tsarin haɗin hoto na Stable Diffusion, amma yana ba da damar samar da abun ciki na gani mai girma uku, kamar hotunan panoramic mai siffar zobe wanda za'a iya kallo a yanayin 360-digiri. A gefe mai amfani, ana iya amfani da samfurin a cikin wasanni da tsarin gaskiya mai kama da juna don ƙirƙirar yanayi mai girma uku.
An horar da ƙirar LDM3D akan babban kwamfuta na Intel AI tare da na'urori masu sarrafa Intel® Xeon® da Intel® Habana Gaudi® AI accelerators.
Ga masu sha'awar aikin, ya kamata su san hakan ana ba da samfurin shirye-shiryen amfani don saukewa kyauta don tsarin koyon injin, wanda ana iya amfani da PyTorch da lambar da aka tsara don samar da hotuna ta amfani da samfura daga aikin Stable Diffusion.
daraja ambata fiye da samfurin ana rarraba ƙarƙashin lasisin izini Ƙirƙirar ML OpenRAIL-M, wanda damar kasuwanci amfani. Rarraba ƙarƙashin buɗaɗɗen lasisi yana ba masu bincike da masu haɓakawa masu sha'awar haɓaka samfuri gwargwadon buƙatunsu da haɓaka shi don ƙwararrun aikace-aikace.
A ƙarshe, idan kuna sha'awar ƙarin sani game da shi, zaku iya tuntuɓar cikakkun bayanai A cikin mahaɗin mai zuwa.