Novel View Synthesis Quiz

Answers and explanations appear as you take the quiz.

1. Given photos of a scene, what does novel view synthesis try to produce?
   • To compress images into smaller files
   • To translate image captions into other languages
   • To detect objects in a single photo
   • To render a scene from camera viewpoints that were not originally captured
2. What does a Neural Radiance Field (NeRF) map a 3D position and viewing direction to?
   • A text description
   • Color (radiance) and volume density
   • A camera serial number
   • A compressed audio signal
3. How does NeRF render a pixel for a new view?
   • By copying the nearest training image directly
   • By averaging all pixels in the dataset
   • By volumetric ray marching: sampling points along the ray and integrating color and density
   • By running optical character recognition
4. What distinguishes 3D Gaussian Splatting's scene representation from an original NeRF-style per-ray network evaluation?
   • Explicit 3D Gaussian primitives rendered through a differentiable rasterization approach
   • A representation containing no scene parameters or image observations
   • Only a table of audio phonemes
   • A method that copies a single input photo for every viewpoint
5. Why can NVS methods reproduce reflections and shiny highlights?
   • They ignore lighting entirely
   • They model view-dependent appearance, so color can change with viewing direction
   • They only render in grayscale
   • They use GPS coordinates
6. What representation technique is central to Instant-NGP's acceleration of neural graphics tasks?
   • Removing all input observations from training
   • Converting the scene into a spreadsheet of filenames
   • A guarantee of the same training time for every scene
   • A multiresolution hash encoding
7. What supplies the photometric training target in original NeRF scene optimization?
   • Observed pixel colors in the training images, compared with rendered colors
   • A required ground-truth 3D mesh for every scene
   • Only the scene's written caption
   • Only the final image's file size
8. Why did original NeRF use positional encoding on its coordinate inputs?
   • To identify the camera's serial number
   • To supply surveyed coordinates without measurements
   • To help the network represent high-frequency spatial and directional variation
   • To eliminate every need for a radiance-field network