Pix2Pix Image-to-Image Translation Quiz

Answers and explanations appear as you take the quiz.

1. What kind of training data does Pix2Pix require?
   • Paired input-output image examples
   • No data at all
   • Only text captions
   • Audio recordings
2. What generator architecture does Pix2Pix use?
   • A plain fully connected network
   • A U-Net with skip connections
   • A recurrent network
   • A decision tree
3. What does the PatchGAN discriminator in Pix2Pix evaluate?
   • The entire image as one global score
   • The image file size
   • Realism of small local image patches
   • The training time
4. Why does Pix2Pix add an L1 loss alongside the adversarial loss?
   • To slow down training on purpose
   • To remove the discriminator
   • To generate random noise
   • To anchor pixelwise and low-frequency fidelity while the adversarial loss encourages sharper texture
5. Why are the U-Net skip connections especially helpful for translation tasks?
   • They pass shared structure like edges directly across the network
   • They delete unnecessary layers
   • They convert images to text
   • They reduce the dataset size
6. Which model extended Pix2Pix to learn from unpaired image collections?
   • ResNet
   • CycleGAN
   • BERT
   • Word2Vec
7. Why can a PatchGAN discriminator encourage sharper local image texture?
   • It predicts the input labels directly
   • It replaces the generator’s U-Net
   • It judges realism in small image patches rather than only scoring the whole image
   • It computes only global color histograms
8. In Pix2Pix, which part of the objective most directly anchors an output to its paired target image?
   • The learning-rate warmup
   • The GAN’s random seed
   • The PatchGAN receptive-field count
   • The L1 pixel-difference loss