Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Idanwo Autoencoders masked

Review MAE’s high-mask image reconstruction task, its asymmetric encoder-decoder design, and the paper’s evaluated pretraining setup.

Jẹmọ itọsọna awọn ọnaAwọn koodu ti ko boju mu
Ibeere 1 ti 8

Kini iṣẹ-ṣiṣe ti ara ẹni ti o ni abojuto mojuto ni Autoencoder Masked?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini iṣẹ-ṣiṣe ti ara ẹni ti o ni abojuto mojuto ni Autoencoder Masked?
    • Ṣiṣe atunṣe awọn piksẹli ti awọn abulẹ ti o farapamọ
    • Pipin aworan kọọkan si ẹka ti o ni aami
    • Ṣe asọtẹlẹ fireemu atẹle ni ọna fidio kan
    • Baramu aworan kan si akọle ọrọ rẹ
  2. Ni aijọju kini ida ti awọn abulẹ aworan ṣe MAE atilẹba ṣe boju-boju jade bi?
    • Nipa 15%
    • Nipa 75%
    • Nipa 35%
    • Nipa 95%
  3. Kilode ti ilana koodu koodu MAE nikan ni awọn abulẹ ti o han (ti ko boju mu)?
    • Nitori awọn abulẹ boju-boju ko ni awọ ti o wulo
    • Lati ṣe idiwọ awoṣe lati rii awọn aworan gidi lailai
    • Lati jẹ ki ikẹkọ yiyara ati iranti diẹ sii daradara
    • Nitoripe oluyipada ko le ka awọn koodu ifidipo
  4. Kini yoo ṣẹlẹ si decoder lẹhin ikẹkọ iṣaaju MAE ti pari?
    • O di akọkọ classifier
    • O ti wa ni didi ati lo lati ṣe ipilẹṣẹ awọn aworan tuntun
    • O rọpo kooduopo patapata
    • O ti wa ni asonu; a ti lo kooduopo fun awọn iṣẹ-ṣiṣe ni isalẹ
  5. Iṣẹ ipadanu wo ni MAE maa n lo fun atunkọ?
    • Itumọ aṣiṣe onigun mẹrin lori awọn piksẹli-patch ti o boju-boju
    • Cross-entropy lori awọn akole kilasi
    • Itansan InfoNCE pipadanu
    • Adversarial discriminator pipadanu
  6. Apẹrẹ 'asymmetric' ti MAE tọka si kini?
    • Awọn titobi aworan oriṣiriṣi fun titẹ sii ati iṣẹjade
    • Ayipada koodu ti o wuwo so pọ pẹlu oluyipada iwuwo fẹẹrẹ kan
    • Lilo awọn iwe data lọtọ meji fun ikẹkọ
    • Masking nikan ni idaji osi ti gbogbo aworan
  7. On which large image dataset did the original MAE paper pretrain its main vision models?
    • MS COCO Captions only
    • LibriSpeech
    • 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