Ikibazo gihuza ibibazo · Hagati Urwego

ESRGAN na GAN Ikibazo Cyibisubizo

Wige ESRGAN's adversarial image-upscaling design, its realptual realism tradeoff, and the limit of ibisobanuro byatanzwe.

Inzira zijyanye no kuyoboraEsrgan na GAN Igisubizo Cyiza
Ikibazo 1 Bya 8

'GAN' muri ESRGAN bivuga iki?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. 'GAN' muri ESRGAN bivuga iki?
    • Umuyoboro rusange wo guhangana
    • Umuyoboro wa Gradient
    • Ikigereranyo cya Gaussian
    • Umuyoboro Witonderwa ku Isi
  2. Niyihe nyubako itanga amashanyarazi ya ESRGAN ikoresha cyane cyane?
    • Ingirabuzimafatizo ndende
    • Ibisigisigi-by-ibisigisigi byuzuye (RRDB)
    • Inzira ya capsule
    • Guhindura kwiyitaho
  3. Kuki abanditsi ba ESRGAN bakuyeho ibisanzwe bisanzwe muri generator?
    • Byadindije imyitozo cyane
    • Yiyongereye kwibuka birenze imipaka
    • Byakunze kumenyekanisha ibihangano biboneka
    • Ntabwo byari bihuye na GAN
  4. Ni ubuhe buryo nyamukuru bwo gucuruza ESRGAN ikora ugereranije na pigiseli-igihombo-gusa?
    • Umuvuduko hejuru yukuri
    • Ibara risobanutse neza
    • Ububiko buke hejuru kurwego rwo hejuru
    • Imyumvire nyayo hejuru ya pigiseli nyayo
  5. Igihombo cya ESRGAN (VGG) gipimirwa he?
    • Kubiranga mbere yo gukora
    • Kuri pigiseli mbisi ya RGB gusa
    • Ku manota ya nyuma y'ivangura
    • Kuri Fourier-yahinduwe
  6. Ni ubuhe bushya ivangura rya ESRGAN ryatangije?
    • Autoencoder gusa
    • Ivangura rishingiye ku gereranya ugereranije nukuri na realism yibinyoma
    • Icyitegererezo cyo gushimangira-kwiga ibihembo
    • Ikonjesha ryateganijwe mbere
  7. Which competition did the ESRGAN paper report winning?
    • The ImageNet classification challenge
    • The Vesuvius Challenge
    • The PIRM 2018 perceptual super-resolution challenge
    • The COCO image-captioning challenge
  8. Why should generated detail from ESRGAN not be treated as recovered photographic evidence?
    • The model records the original sensor data exactly
    • GANs can only enlarge images without changing pixels
    • The output is guaranteed to match the lost high-resolution image
    • The model can synthesize plausible textures that were not present in the low-resolution input