GFPGAN Face Restoration Quiz

Answers and explanations appear as you take the quiz.

1. What pretrained model provides GFPGAN's 'generative facial prior'?
   • StyleGAN2
   • BERT
   • ResNet-50
   • Whisper
2. What is the role of GFPGAN’s CS-SFT layers?
   • They compress the image file
   • They blend prior features with input-specific information to balance realism and fidelity
   • They translate captions
   • They detect objects in the background
3. What does the GFPGAN paper report about inference?
   • It needs exactly 50 optimization passes
   • It performs one pass per image pixel
   • It restores facial details in a single forward pass
   • It iterates until the input becomes a perfect photograph
4. What companion model is often paired with GFPGAN to handle the background?
   • GPT-4
   • YOLO
   • Stable Diffusion XL
   • Real-ESRGAN
5. During training, what does GFPGAN NOT require?
   • Paired high-quality photos of the same specific person
   • Adversarial loss
   • A pretrained face prior
   • Perceptual loss
6. What is a key ethical concern with GFPGAN-style restoration?
   • It uses too little electricity
   • It can invent plausible facial detail that was not in the original
   • It only works on cartoons
   • It deletes the original file automatically
7. What is the full name of GFPGAN’s CS-SFT component?
   • Color-Space Semantic Face Tracking
   • Convolutional Style-Filter Training
   • Channel-Split Spatial Feature Transform
   • Channel-Specific Super-Resolution Fourier Transform
8. In blind face restoration, what does “blind” mean?
   • The face is hidden behind an object
   • The system only processes black-and-white images
   • The model never sees an input image
   • The exact degradation process is not known in advance