CodeFormer Robust Face Recovery Quiz

Answers and explanations appear as you take the quiz.

1. How does CodeFormer fundamentally frame the face restoration problem?
   • As discrete code prediction from a learned codebook
   • As continuous pixel regression
   • As audio transcription
   • As text summarization
2. What component predicts which codebook entries to use?
   • A random number generator
   • A Transformer
   • A decision tree
   • A Fourier transform
3. Why can CodeFormer’s discrete codebook help with severely degraded faces?
   • It converts images to grayscale
   • It removes all color information
   • It constrains predictions to a learned vocabulary of facial visual atoms
   • It requires the input to be pristine
4. What does CodeFormer’s controllable feature transformation setting adjust?
   • Image resolution only
   • File format
   • The number of faces detected
   • The trade-off between fidelity to the input and restored visual quality
5. In which venue was CodeFormer published?
   • NeurIPS 2022
   • SIGGRAPH 1999
   • A cooking magazine
   • It was never published
6. What technique does the codebook training derive from?
   • Linear regression
   • A VQGAN-style vector-quantized model
   • A relational database
   • A spreadsheet macro
7. What two needs does the CodeFormer paper identify for blind face restoration guidance?
   • Changing image format and compressing files
   • Detecting camera brand and lens type
   • Improving the degraded-input mapping and complementing details lost from the input
   • Removing color and resizing the image
8. What kinds of test data did the CodeFormer paper use in its reported experiments?
   • Only hand-drawn illustrations
   • Only pristine studio portraits
   • Audio and text datasets
   • Synthetic and real-world degraded faces