CycleGAN Unpaired Translation Quiz

Answers and explanations appear as you take the quiz.

1. What key problem does CycleGAN solve that earlier methods like pix2pix could not?
   • It translates images without needing matched before-and-after pairs
   • It runs entirely on a mobile CPU
   • It eliminates the need for any training data
   • It only works on black-and-white images
2. What does the cycle-consistency loss enforce?
   • That all output images are grayscale
   • That translating an image to the other domain and back recovers the original
   • That the discriminator always wins
   • That generators share the same weights
3. How many generators does a standard CycleGAN use?
   • One
   • Four
   • Two
   • Three
4. What type of discriminator does CycleGAN typically use?
   • A single sigmoid on the whole image
   • A pretrained ImageNet classifier
   • No discriminator at all
   • A PatchGAN that classifies image patches
5. Why is the identity loss sometimes added in CycleGAN?
   • To help preserve color when an input already matches the target domain
   • To make training faster
   • To remove the need for discriminators
   • To convert images to sketches
6. Which of these is a classic CycleGAN demonstration?
   • Generating text from speech
   • Turning horse photos into zebra photos
   • Transcribing handwriting
   • Compressing video files
7. Why does a standard CycleGAN use one discriminator for each image domain?
   • Each one estimates the image’s text caption
   • One discriminator identifies the source dataset and the other the author
   • Each discriminator judges whether generated images look realistic in its target domain
   • They replace the two generators
8. How can CycleGAN support synthetic-to-real adaptation for autonomous-driving vision?
   • Generate steering labels from vehicle sounds
   • Replace all road images with segmentation text
   • Identify a driver from a voice recording
   • Translate simulator images toward the visual style of real road scenes without paired matches