Wata ƙungiyar masu bincike daga Jami'ar Tel Aviv ta buɗe lambar tushe da ke da alaƙa da tsarin ilmantarwa na injin MDM (Motion Diffusion Model), wanda ke ba da damar samar da motsin ɗan adam na gaske. An rubuta lambar a cikin Python ta amfani da tsarin PyTorch kuma an rarraba a ƙarƙashin lasisin MIT. Don gudanar da gwaje-gwaje, zaku iya amfani da samfuran shirye-shiryen duka kuma ku horar da samfuran da kanku ta amfani da rubutun da aka tsara, alal misali, ta amfani da tarin HumanML3D na hotunan ɗan adam mai girma uku. Don horar da tsarin, ana buƙatar GPU tare da tallafin CUDA.
Amfani da damar al'ada don raya motsin ɗan adam yana da wahala saboda rikice-rikicen da ke tattare da ɗimbin ƙungiyoyi masu yuwuwa da wahalar siffanta su a hukumance, da kuma babban azancin fahimtar ɗan adam ga ƙungiyoyin da ba na dabi'a ba. Ƙoƙarin da aka yi a baya na yin amfani da samfuran koyo na injin ƙira sun sami matsala tare da inganci da ƙayyadaddun bayyanawa.
Tsarin da aka gabatar yana ƙoƙarin yin amfani da samfuran watsawa don samar da motsi, waɗanda a zahiri sun fi dacewa don kwaikwayon motsin ɗan adam, amma ba tare da lahani ba, kamar manyan buƙatun ƙididdiga da sarkar sarrafawa. Don rage gazawar samfuran watsawa, MDM tana amfani da hanyar sadarwa ta jijiyar canji da tsinkayar samfur maimakon tsinkayar amo a kowane mataki, yana sauƙaƙa don hana abubuwan da ba su da kyau kamar asarar hulɗar ƙasa da ƙafa.
Don sarrafa tsarawa, yana yiwuwa a yi amfani da bayanin rubutu na wani aiki a cikin harshen halitta (misali, "mutum yana tafiya gaba kuma ya lanƙwasa ƙasa don ɗaukar wani abu daga ƙasa") ko amfani da daidaitattun ayyuka kamar "gudu" da " tsalle." Hakanan za'a iya amfani da tsarin don gyara ƙungiyoyi da cika bayanan da suka ɓace. Masu binciken sun gudanar da gwaji inda aka tambayi mahalarta don zaɓar mafi kyawun sakamako daga zaɓuɓɓuka da yawa - a cikin 42% na lokuta, mutane sun fi son haɗakar ƙungiyoyi akan na ainihi.
source: budenet.ru