Masked Autoencoders
Masked Autoencoders (MAE) hanya ce mai kulawa da kai wanda ke koyar da ƙirar hangen nesa don sake gina hotuna bayan an ɓoye yawancin hoton.
Dubawa
By learning to fill in the blanks, the model builds rich visual understanding without any human labels.
Zurfafa nutsewa
Masked Autoencoders, wanda Kaiming He da abokan aiki suka gabatar a Meta AI a cikin 2021, suna ɗaukar hoto, raba shi zuwa ƙananan faci, kuma ba da gangan ɓoye babban juzu'in su ba, yawanci 75%. Mai canza hangen nesa yana aiwatar da facin da ake iya gani kawai, yayin da na'ura mai sauƙi yana ƙoƙarin sake gina ainihin pixels na waɗanda suka ɓace. Saboda abubuwa da yawa suna ɓoye, ƙirar ba za ta iya kwafin pixels na kusa ba kawai kuma dole ne su koyi tsari mai ma'ana, kamar siffofi da sassan abubuwa. Mai rikodin rikodin tsallake facin abin rufe fuska yana sa horo cikin sauri da ingantaccen ƙwaƙwalwar ajiya. Bayan riga-kafi, ana jefar da mai rikodin rikodin kuma mai rikodin yana canjawa da ƙarfi zuwa rarrabuwa, ganowa, da ayyukan rarrabawa.
Fahimtar Fasaha
Dabarar maɓalli shine asymmetry: mai ɗaukar hoto mai nauyi yana ganin kashi 25% na facin da ba a rufe shi ba, yayin da ƙaramin dikodi ke sake gina sauran. Faci-faci an daidaita su, an saka su a layi, kuma an ba su rikodi na matsayi. Asarar sake ginawa ita ce kuskuren ƙididdigewa mai murabba'i da aka ƙididdige shi kawai akan faci da aka rufe, yawanci akan ƙimar pixel daidaitacce. Matsakaicin abin rufe fuska yana tilasta koyo na ma'anar maimakon ƙaramin matakin tsaka-tsaki, da tsallake alamomin rufe fuska a cikin yankan ramin ƙididdigewa da sarrafa cikakken hoto.
Dabarun Tasiri
Gudu da sikelin
Kayayyakin AI na iya sarrafa aiki da bincike, ganowa, da ayyuka masu alama a sikelin.
Gina zaɓuɓɓuka
Ƙungiyoyin ƙirƙira za su iya samar da ra'ayoyi cikin sauri tare da ƙarancin bita da hannu.
Ƙungiya da aikin aiki
Ayyuka na iya amfani da siginar hoto da bidiyo waɗanda a baya suke da wahalar aiwatarwa.
Makomar Masked Autoencoders
Sake gina abin rufe fuska-style MAE yana zama tsoho girke-girke na pretraining a duk hanyoyin. Masu bincike suna mika shi zuwa bidiyo (boye kubes na sararin samaniya), na'urar kallon sauti, sikanin likitanci, da hotunan tauraron dan adam, inda alamomin ba su da yawa kuma masu tsada. Yi tsammanin haɗuwa mai ƙarfi tare da harshe don ƙirar tushen tushe na multimodal, mafi inganci na'urar dikodi, da abin rufe fuska mai daidaitawa waɗanda ke keɓance yankuna masu ba da labari. Yayin da ƙididdigewa ke girma, horarwar da abin rufe fuska akan ɗimbin hotuna marasa lakabi ya kamata ya ci gaba da haɓaka daidaiton ƙasa tare da rage dogaro ga bayanin ɗan adam mai tsada.
Aiwatar da Gaskiyar Duniya
Gabatar da Mai Canja Hankali akan miliyoyin hotuna marasa lakabi, sannan kuma daidaita shi don rarraba ImageNet tare da daidaito mai ƙarfi.
Fasalolin koyo daga sikanin likitanci mara lakabi (X-ray, MRIs) inda bayanin ƙwararru ke da tsada kuma yana da iyaka.
Daidaita hanyar zuwa bidiyo ta hanyar rufe faci na lokacin sararin samaniya don horar da ƙirar aikin-gane (VideoMAE)
Horarwa akan tauraron dan adam da hotunan sararin sama don tallafawa taswirar amfani da ƙasa da kuma canza ganowa ba tare da alamun hannu ba.
Hatsari & Tsare-tsare
Haƙƙoƙin hoto da yarda na iya zama haxarin doka idan ba a fayyace ba.
Ayyukan samfuri na iya bambanta a ko'ina cikin haske, ƙididdiga, da mahalli.
Ƙarya tabbataccen ƙila ba za a iya lura da shi ba sai dai idan an kula da ƙofofin amincewa.
Taswirar Hanya
Ƙayyade ma'auni na karɓa don daidaito, tunowa, da farashi na kuskure.
Gwada tare da bayanan da suka dace da ainihin yanayin samarwa.
Ƙara bita na ɗan adam don ƙarancin amincewa ko tsinkaya mai tasiri.
Bi diddigin ƙirar ƙira kuma sake ingantawa bayan canje-canjen kamara ko saitin bayanai.
Ci gaba da Bincike
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 Masked Autoencoders 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
Jagora na gaba
Muse Masked Generative Hoto
Tambayoyin da ake yawan yi
What is Masked Autoencoders?
Masked Autoencoders (MAE) hanya ce mai kulawa da kai wanda ke koyar da ƙirar hangen nesa don sake gina hotuna bayan an ɓoye yawancin hoton. Ta hanyar koyon cike guraben, ƙirar tana gina wadataccen fahimtar gani ba tare da kowane tambarin ɗan adam ba.
Menene ainihin aikin kulawa da kai a cikin Maɓallin Autoencoder?
MAE yana ɓoye yawancin facin hoto kuma yana horar da ƙirar don sake gina pixels da suka ɓace, ba buƙatar alamar ɗan adam ba.
Kusan wane yanki na facin hoto ne ainihin MAE ke rufewa?
Asalin MAE masks kusan kashi 75% na faci, babban rabo wanda ke tilasta ƙirar don koyon tsari mai ma'ana maimakon kwafin makwabta.
Me yasa MAE encoder ke aiwatar da faci na bayyane (wanda ba a rufe) kawai?
Tsallake alamun rufe fuska a cikin rikodi yana rage ƙididdigewa da ƙwaƙwalwa sosai, tunda yana ɗaukar ɗan ƙaramin faci ne kawai.
Me zai faru da mai gyara bayan an gama horaswar MAE?
Ana buƙatar na'urar mai sauƙi mai sauƙi kawai don sake ginawa yayin horarwa; bayan haka mai rikodin koyo yana canjawa zuwa ayyuka kamar ganowa.
Wanne hasara ce MAE ke amfani da ita don sake ginawa?
MAE yana rage ma'anar kuskure mai murabba'i tsakanin kimar pixel da aka annabta da na gaskiya, wanda aka lissafta akan facin da aka rufe kawai.