quiz bu lëkkale ak tegtal · Diggante Niveau

Google

Xoolaat ni Imagen di doxalee ci diffusion ci bind, gëna suqali nataal yi, ba noppi jëfandikoo tegtal ak tënk ci diiru defar.

Yooni njiit yu jëm ciNataalu Google
Laaj 1 de 8

Ban xeetu xeetu defar la Imagen tabax ci kaw?

Xoolaatal laaj bu nekk8 laaj

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

  1. Ban xeetu xeetu defar la Imagen tabax ci kaw?
    • ab xeetu diffusion
    • Reseau buy xeex (GAN)
    • Autoencodeur bu wuute kese
    • Renderer bu sukkandiko ci sàrt
  2. Benn gis-gis bu am solo ci këyitu Imagen bi njëkk mooy scaling ban composant moo gëna yokk njariñ yi?
    • Reseau diffusion nataal
    • Encodeur de text congélé (modèle de langue)
    • Dayo tàggat yaram
    • limu chaine yu melo yi
  3. Which improvement did Google announce for Imagen 4 in the Gemini app?
    • Denc fichier yi ci anamu PNG
    • Defar ay audio
    • Improved text rendering in generated images
    • Komprime nataal yi
  4. Architecture bu njëkk bu Imagen dafa jëfandikoo cascade bu tàmbali ci defar nataal ci ban dayo?
    • 1024x1024
    • 512x512
    • 4096x4096
    • 64x64
  5. What did Google DeepMind say SynthID does for images generated with Imagen on Vertex AI?
    • Embeds an imperceptible watermark in image pixels to help identify likely Imagen output
    • Converts all images to a new file format
    • Provides a GPU chip for image training
    • Translates the prompt language
  6. Ban platform la defarkat yi mëna jaar ngir jëfandikoo Imagen ngir seeni aplikaasioŋ?
    • Fiche Google
    • IA
    • Google Maps
    • Jumtukaayi Developpement Chrome
  7. What issue does dynamic thresholding address in Imagen's sampling process?
    • The model having too few text tokens in its prompt
    • The training set containing too many image resolutions
    • Pixel values becoming oversaturated when classifier-free guidance is high
    • The tokenizer splitting a word into subwords
  8. What does classifier-free guidance do during Imagen sampling?
    • Selects a class label after the image is complete
    • Upscales the image without conditioning on its prompt
    • Replaces the noise-prediction model with a fixed image lookup
    • Combines conditional and unconditioned predictions to steer generation toward the prompt