MaskGIT Parallel Token Decoding Quiz

Answers and explanations appear as you take the quiz.

1. How does MaskGIT decode image tokens differently from VQGAN's transformer?
   • It predicts tokens in parallel and refines over a few steps
   • It predicts one token at a time in raster order
   • It uses no transformer
   • It generates only text
2. What training objective does MaskGIT borrow from language models?
   • Next-word-only prediction
   • Masked token prediction, like BERT
   • Image classification
   • Reinforcement learning from human feedback
3. During generation, which tokens does MaskGIT commit at each iteration?
   • The lowest-confidence ones
   • Random tokens
   • The highest-confidence predictions
   • All tokens at once on step one
4. In the original MaskGIT generation examples, roughly how many refinement iterations were often used?
   • Exactly one iteration for every image and task
   • Hundreds of sequential token predictions
   • A fixed number that cannot be changed by the user
   • About 8 to 12 in those reported configurations
5. What schedule controls how many tokens MaskGIT reveals each step?
   • A cosine masking schedule
   • A linear-only rule revealing all at once
   • No schedule; it's random
   • A fixed one token per step
6. Why does bidirectional attention help MaskGIT?
   • It hides the image from the model
   • It lets each token see the whole partial image for better context
   • It only looks at earlier tokens
   • It removes the need for any tokens
7. What token grid does MaskGIT begin with when generating an image?
   • A fully decoded image that only needs color correction
   • A left-to-right stream with no masked positions
   • A grid with every image token masked
   • A grid of text embeddings from the prompt only
8. How can MaskGIT fill a masked region during image inpainting?
   • Discard the observed image and generate an unrelated full image
   • Use only a left-to-right text decoder with no image tokens
   • Replace every visible token with a random codebook entry
   • Keep observed image tokens as context and iteratively predict the masked positions