Ihe mkpuchi Autoencoders
Masked Autoencoders (MAE) bụ usoro a na-ahụ maka onwe ya nke na-akụzi ihe ngosi ọhụụ iji wughachi ihe oyiyi mgbe ezorori ọtụtụ foto ahụ.
Nchịkọta
By learning to fill in the blanks, the model builds rich visual understanding without any human labels.
Ime miri emi
Masked Autoencoders, nke Kaiming He webatara ya na ndị ọrụ ibe ya na Meta AI na 2021, see foto, kewaa ya na obere patches, wee zoo obere akụkụ dị ukwuu n'ime ha na-enweghị usoro, mgbe mgbe 75%. Ihe ntụgharị ntụgharị ọhụụ na-arụ ọrụ naanị patches a na-ahụ anya, ebe ihe ngbanwe dị fechaa na-agbalị imegharị pikselụ mbụ nke ndị na-efu efu. N'ihi na ezoro ezo nke ukwuu, ihe nlere ahụ enweghị ike iṅomi naanị pikselụ dị nso na ọ ga-amụrịrị usoro bara uru, dị ka ụdị na akụkụ ihe. Ihe ngbanwe na-awụpụ patches nkpuchi na-eme ka ọzụzụ dị ngwa na ebe nchekwa rụọ ọrụ nke ọma. Mgbe emechara ọzụzụ mbụ, a ga-atụfu ihe ndozi ahụ yana ihe ngbanwe ahụ na-ebufe ike na nhazi, nchọpụta na ọrụ nkewa.
Nghọta nka nka
Isi aghụghọ bụ asymmetry: ihe mkpuchi dị arọ na-ahụ naanị 25% nke patches na-enweghị mkpuchi, ebe obere decoder na-emegharị ndị ọzọ. A na-agbajikwa patches, tinye n'ahịrị n'ahịrị, ma nye ya ntinye ọnọdụ. Mfu nrụgharị ahụ bụ njehie gbara mkpị gbakọrọ naanị na patches kpuchiri ekpuchi, na-abụkarị na ụkpụrụ pikselụ ahaziri ahazi. Oke nkpuchi dị elu na-amanye mmụta mmụta sayensị kama itinye aka n'ọkwa dị ala, na ịwụpụ akara mkpuchi mkpuchi na mbelata ihe ngbanwe na-agbakọ nke ọma na nhazi onyonyo zuru oke.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Visual AI nwere ike megharịa nyocha, nchọpụta na mkpado ọrụ n'ọtụtụ.
Mee nhọrọ
Otu ndị na-emepụta ihe nwere ike imepụta echiche ngwa ngwa site na ngbanwe akwụkwọ ntuziaka ole na ole.
Team na usoro ọrụ
Ọrụ nwere ike iji onyonyo na akara vidiyo siri ike ịhazi.
Ọdịnihu nke mkpuchi Autoencoders
Nrụgharị ihe mkpuchi ụdị MAE na-aghọ usoro nkuzi izizi n'usoro n'usoro. Ndị na-eme nchọpụta na-agbatị ya na vidiyo (na-ezobe cubes spacetime), ihe nkiri onyonyo, nyocha ahụike, na onyonyo satịlaịtị, ebe akara ngosi dị ụkọ ma dị oke ọnụ. Na-atụ anya njikọ siri ike na asụsụ maka ụdị ntọala multimodal, ndị na-emepụta ihe na-arụ ọrụ nke ọma, na ihe nkpuchi na-agbanwe agbanwe nke na-elekwasị anya na mpaghara ndị na-enye ozi. Ka mgbakọ na-eto eto, ọzụzụ nkpuchi kpuchiri ekpuchi na nnukwu mkpokọta onyonyo a na-edeghị aha kwesịrị ịnọgide na-emeziwanye izi ezi nke ala na-ebelata ịdabere na nkọwa mmadụ dị oke ọnụ.
Mmejuputa n'ezie n'ụwa
Ịkwalite ihe ntụgharị ọhụụ na nde foto ndị na-enweghị aha, wee dozie ya nke ọma maka nhazi nke ImageNet n'ụzọ ziri ezi.
Atụmatụ mmụta site na nyocha ahụike na-enweghị akara (X-ray, MRI) ebe nkọwa ndị ọkachamara dị oke ọnụ na oke.
Ịmegharị usoro a ka ọ bụrụ vidiyo site na ikpuchi patches spacetime ka ọ na-azụ ụdị ọrụ-amata (VideoMAE)
Ịmalite ọzụzụ na satịlaịtị na onyonyo ikuku iji kwado nkewa iji ala wee gbanwee nchọpụta na-enweghị akara akwụkwọ ntuziaka.
Ihe ize ndụ & okporo ụzọ nche
Ikike onyonyo na nkwenye nwere ike bụrụ ihe egwu dị n'iwu ma ọ bụrụ na edoghị anya.
Ọrụ nlereanya nwere ike ịdịgasị iche n'ofe ọkụ, igwe mmadụ, na gburugburu.
Enwere ike ghara ịhụ ihe dị mma ma ọ bụrụ na enyochaghị oke ntụkwasị obi.
Map mmejuputa
Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.
Nwalee na data dabara na ọnọdụ mmepụta n'ezie.
Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.
Sochie ihe nlere anya wee megharịa ka emechara mgbanwe igwefoto ma ọ bụ dataset.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Muse Masked Generative Imaging
Ajụjụ a na-ajụkarị
What is Masked Autoencoders?
Masked Autoencoders (MAE) bụ usoro a na-ahụ maka onwe ya nke na-akụzi ihe ngosi ọhụụ iji wughachi ihe oyiyi mgbe ezorori ọtụtụ foto ahụ. Site n'ịmụta imeju oghere, ihe nlereanya ahụ na-ewulite nghọta ọhụụ bara ụba na-enweghị akara mmadụ ọ bụla.
Gịnị bụ isi ọrụ a na-ahụ maka onwe ya na nkpuchi Autoencoder?
MAE na-ezochi ọtụtụ patches oyiyi ma zụọ ihe nlereanya iji wughachi pikselụ efu, na-achọghị akara mmadụ.
Kedu obere akụkụ nke patches oyiyi ka MAE mbụ na-ekpuchikarị?
Ihe mkpuchi MAE mbụ dị ihe dị ka 75% nke patches, oke dị elu nke na-amanye ihe nlereanya ahụ ịmụta usoro bara uru kama iṅomi ndị agbata obi.
Kedu ihe kpatara na koodu MAE na-ahazi naanị patches a na-ahụ anya (enweghị mkpuchi)?
Ịwụfe akara nkpuchi kpuchie na koodu ntinye na-ebelata mkpokọta na ebe nchekwa, ebe ọ na-ejikwa naanị obere mperi nke patches.
Kedu ihe na-eme ihe ngbanwe mgbe ọzụzụ MAE mechara?
Ihe ndozi dị fechaa bụ naanị maka nwughari n'oge ọzụzụ izizi; emesia ihe ngbanwe a mụtara na-ebufe na ọrụ dịka nchọpụta.
Kedu ọrụ mfu ka MAE na-ejikarị emezigharị?
MAE na-ebelata mperi pụtara square n'etiti ụkpụrụ pikselụ buru amụma na nke ezi, gbakọọ naanị na patches kpuchie.