DUSt3R Dense 3D Reconstruction Quiz

Answers and explanations appear as you take the quiz.

1. What key piece of information does DUSt3R NOT require as input, unlike classic structure-from-motion?
   • Known camera poses and calibration
   • Any images at all
   • A graphics card
   • Color in the images
2. What is the central output that DUSt3R's network regresses for each image?
   • A text caption
   • A pointmap of per-pixel 3D coordinates
   • A bounding box
   • A segmentation mask
3. In a DUSt3R image pair, what coordinate frame are both pointmaps expressed in?
   • A random global frame
   • World GPS coordinates
   • The coordinate frame of the first camera
   • Each image's own separate frame
4. How can DUSt3R produce camera poses and pixel matches without receiving them as known input?
   • A separate calibration rig must provide every pose
   • Human annotators label each pixel correspondence
   • The model cannot produce either output
   • They can be recovered downstream from the predicted pointmaps and their alignment
5. How does DUSt3R handle more than two input images?
   • It performs a global alignment stitching pairwise pointmaps together
   • It ignores all but two
   • It averages the images into one photo
   • It requires a separate network per image
6. Which follow-up method extends DUSt3R with stronger dense matching?
   • ResNet
   • MASt3R
   • Stable Diffusion
   • NeRF
7. What additional per-pixel output accompanies DUSt3R pointmaps?
   • A speech transcript
   • A musical pitch label
   • A confidence map
   • A JPEG compression code
8. What model architecture does DUSt3R use to process image pairs?
   • A rule-based camera calibration script only
   • A convolutional audio vocoder
   • A database query engine
   • Transformer encoders and decoders with cross-view processing