Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Onitẹsiwaju Dagba ti GANs adanwo

Test coarse-to-fine GAN training, symmetric growth, alpha fade-in, batch-variation features, equalized learning rates, and later fixed architectures.

Jẹmọ itọsọna awọn ọnaIdagba ilọsiwaju ti GANs
Ibeere 1 ti 8

Bawo ni GAN ti o dagba ni ilọsiwaju bẹrẹ ikẹkọ?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Bawo ni GAN ti o dagba ni ilọsiwaju bẹrẹ ikẹkọ?
    • Ni ipinnu ibi-afẹde ni kikun lẹsẹkẹsẹ
    • Ni ipinnu kekere pupọ bi awọn piksẹli 4x4
    • Nipa pretraining nikan ni discriminator
    • Pẹlu itọka ọrọ
  2. Kini anfani akọkọ ti ipinnu idagbasoke diẹdiẹ?
    • O jẹ ki ikẹkọ jẹ iduroṣinṣin ati yiyara ju ikẹkọ ni ipinnu giga taara
    • O yọ awọn nilo fun a monomono
    • O ṣe imukuro ibeere dataset kuro
    • O ṣe onigbọwọ awọn ohun-ọṣọ odo
  3. Nigba ti titun kan ti o ga-o ga Layer ti wa ni afikun, bawo ni a ṣe o?
    • Lẹsẹkẹsẹ ni kikun agbara
    • Nikan ni discriminator, kò monomono
    • Díẹ̀rọ̀ rọ̀ sínú rẹ̀ pẹ̀lú ìwọ̀n ìdàpọ̀ (Alpha) tí ó gòkè láti 0 sí 1
    • Nipa atunkọ nẹtiwọki lati ibere
  4. Kilode ti a fi kun awọn fẹlẹfẹlẹ si mejeji monomono ati iyasoto?
    • Nitori monomono nikan ko le gbe awọ jade
    • Lati dinku iwọn dataset
    • Nikan fun awọn idi iworan
    • Lati jẹ ki awọn nẹtiwọọki mejeeji jẹ iwọntunwọnsi ati iwọntunwọnsi bi ipinnu n pọ si
  5. Kini idi ti Layer iyapa boṣewa minibatch ti a ṣafihan ni ProGAN?
    • Lati gbe aworan ikẹhin soke
    • Lati compress awoṣe
    • Lati mu iyatọ iṣẹjade pọ si ati irẹwẹsi ipo iṣubu
    • Lati fi aami si awọn aworan ikẹkọ
  6. Idagbasoke nigbamii wo ni o dinku iwulo fun idagbasoke ti o ni ilọsiwaju kedere?
    • Awọn kiikan ti awọn discriminator
    • StyleGAN2 ti n ṣafihan faaji ti o wa titi pẹlu awọn asopọ fo / iyokù le baamu didara
    • Yipada si awọn aworan grẹy
    • Yiyọ awọn wiwaba fekito patapata
  7. At alpha equal to zero during a ProGAN generator fade-in, which path supplies the blended output?
    • Only the newly added block at full strength
    • An unrelated pretrained text encoder
    • A random photograph taken from the test set
    • The previous lower-resolution path after upsampling
  8. What is the original ProGAN equalized-learning-rate method intended to balance?
    • The effective scale of learning across layers through dynamic weight scaling
    • The number of captions assigned to each face image
    • The timestamp of every generated training sample
    • The count of output file formats across batches