Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Kediska Autoencoders waji-xidhan

Dib u eeg MAE shaqadiisa dib-u-dhiska sawirka-maaskar-sare, nashqad-qodeeyaha-qalabeeyaha asymmetric-ga ah, iyo diyaarinta tababarka kahor ee warqadda la qiimeeyay.

Wadooyinka haga ee la xidhiidhaAutoencoders-ka daboolan
Su'aal 1 ee 8

Waa maxay hawsha aasaasiga ah ee is-maamulida ee ku jirta Autoencoder waji-xidhan?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Waa maxay hawsha aasaasiga ah ee is-maamulida ee ku jirta Autoencoder waji-xidhan?
    • Dib u dhiska pixels ee balastar ee qarsoonaa
    • Kala soocida sawir kasta oo loo qaybiyo qayb calaamadeysan
    • Saadaasha qaabka xiga ee isku xigxiga fiidyaha
    • Ku habboon sawirka qoraalka qoraalka ah
  2. Qiyaastii waa maxay jajabka dhejiska sawirka ee asalka ah ee MAE caadi ahaan maaskaro?
    • Qiyaastii 15%
    • Ilaa 75%
    • Ilaa 35%
    • Ilaa 95%
  3. Waa maxay sababta uu MAE codeeyaha u habeeyo oo kaliya balastar muuqda (aan qarinayn)?
    • Sababtoo ah balastarrada la daboolay kuma jiraan midab faa'iido leh
    • Si looga hortago in moodelku waligiis arko sawirro dhab ah
    • Si aad tababarka uga dhigto mid dhakhso badan oo xusuus badan leh
    • Sababtoo ah codeeyaha ma akhriyi karo codaynta booska
  4. Maxaa ku dhacaya codeeyaha ka dib marka la dhammeeyo tababarka hore ee MAE?
    • Waxay noqotaa kala soocida ugu weyn
    • Waa la qaboojiyey waxaana loo isticmaalaa in lagu soo saaro sawirro cusub
    • Waxay beddeshaa cod-bixiyaha gebi ahaanba
    • Waa la iska tuuraa; encoder-ka waxaa loo isticmaalaa hawlaha hoose
  5. Shaqadee khasaaraha ay MAE caadi ahaan u isticmaashaa dib u dhiska?
    • Khaladka dhexdhexaadka ah ee labajibbaaran ee pixels-patch-ka daboolan
    • Ka gudub-entropy ka sarreeya calaamadaha fasalka
    • Luminta macluumaadka is-khilaafsan
    • Khasaaraha takoorka cadawga
  6. Naqshada 'asymmetric' ee MAE waxa loola jeedaa maxaa?
    • Cabbirrada sawirka kala duwan ee gelinta iyo soo saarista
    • Codeeye culus oo lagu lamaanay qalab-dejiyaha fudud
    • Isticmaalka laba xog-ururin oo kala duwan xagga tababarka
    • Maaskarada oo kaliya qeybta bidix ee sawir kasta
  7. On which large image dataset did the original MAE paper pretrain its main vision models?
    • MS COCO Captions only
    • Hadal-qoraal
    • ImageNet-1K
    • The Netflix Prize ratings set
  8. What does the MAE decoder receive for image patches hidden from the encoder?
    • The original RGB pixels without masking
    • Only class labels from ImageNet
    • A text caption for the image
    • Learned mask tokens with positional embeddings