Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Autoregressive Image Iran adanwo

Learn how autoregressive image models encode images as token sequences and generate them one step at a time from text prompts.

Jẹmọ itọsọna awọn ọnaAutoregressive Image generation
Ibeere 1 ti 8

Kini 'autoregressive' tumọ si ni aaye ti iran aworan?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini 'autoregressive' tumọ si ni aaye ti iran aworan?
    • Ẹya tuntun kọọkan jẹ asọtẹlẹ iloniniye lori awọn eroja ti ipilẹṣẹ tẹlẹ
    • Ẹya tuntun kọọkan jẹ asọtẹlẹ nipa lilo itọ ọrọ nikan
    • Gbogbo aworan ti wa ni denoised ni ẹẹkan lati ID ariwo
    • Awọn piksẹli jẹ ipilẹṣẹ patapata ni ominira ti ara wọn
  2. Bawo ni awọn awoṣe aworan autoregressive ode oni bii DALL-E 1 ṣe aṣoju aworan kan ṣaaju asọtẹlẹ rẹ?
    • Bi awọn kan nikan lemọlemọfún fekito
    • Gẹgẹbi akoj ti o wa titi ti awọn ami iyasọtọ lati inu iwe koodu ti kọ ẹkọ
    • Bi awọn piksẹli 24-bit RGB aise nikan
    • Bi awọn akojọ ti awọn fekito eya ni nitobi
  3. Awọn awoṣe ibẹrẹ wo ni o ṣẹda awọn aworan piksẹli aise kan ni akoko kan?
    • GANs ati VAE
    • Idurosinsin Itankale ati Imagen
    • PixelRNN ati PixelCNN
    • CLIP ati BLIP
  4. Ilana wo ni o ṣe idaniloju Amunawa kan nikan wa si awọn ami ami iṣaaju lakoko iran adaṣe?
    • Jade kuro
    • Ipese deede
    • Rekọja awọn isopọ
    • Idi (masked) akiyesi ara ẹni
  5. Kí ni akọkọ ilowo drawback ti autoregressive image iṣapẹẹrẹ?
    • Iṣapẹẹrẹ jẹ lẹsẹsẹ ati nitorinaa lọra
    • Ko le wa ni iloniniye lori ọrọ
    • Ko le ṣe apẹẹrẹ awọn alaye ti o dara
    • Ko nilo data ikẹkọ
  6. Ilana wo ni o ṣe iyara iran aworan ti o da lori ami-ami nipasẹ sisọ asọtẹlẹ ọpọlọpọ awọn ami ni afiwe?
    • Olukọni fi agbara mu
    • Iyipada ni afiwe ti o boju-boju (fun apẹẹrẹ, MaskGIT/Muse)
    • Idinku gige
    • Ọkan-gbona fifi koodu
  7. How does the original DALL·E model use text and image tokens?
    • Separate unconnected classifiers process text and pixels
    • A diffusion decoder predicts all pixels simultaneously
    • A single autoregressive Transformer models the sequence conditioned on text and image tokens
    • The model uses only a text encoder and no image-token sequence
  8. What scale of image-text pairs did the original DALL·E paper report collecting?
    • About 250 pairs
    • About 25 thousand pairs
    • About 25 billion pairs
    • About 250 million image-text pairs