Differentiable Rendering Quiz

Answers and explanations appear as you take the quiz.

1. What does differentiable rendering let you compute?
   • Gradients of output pixels with respect to scene parameters like geometry and materials
   • A faster ray-tracing algorithm with no gradients
   • Audio from a 3D scene
   • Compressed video files from animations
2. Why are object silhouettes a core difficulty for differentiable rendering?
   • They are always blurry in real photos
   • They cause visibility discontinuities where pixels jump abruptly, giving no useful naive gradient
   • They require more memory than smooth surfaces
   • They change the file format of the output
3. How does soft rasterization address the discontinuity problem?
   • It removes all edges from the scene
   • It increases the resolution until edges disappear
   • It lets triangles contribute a smooth, blurred footprint to nearby pixels
   • It renders only the background
4. What is 'inverse rendering'?
   • Rendering a scene upside down
   • Converting 2D images to audio
   • Rendering only at night
   • Recovering scene parameters (shape, materials, lighting) from images
5. Which modern technique relies on differentiable rendering for training?
   • 3D Gaussian splatting and NeRFs
   • JPEG compression
   • MP3 encoding
   • Spreadsheet pivot tables
6. What is the 'edge sampling' approach (as in redner) designed to capture?
   • The average color of the whole image
   • The boundary term of the rendering integral at discontinuities
   • The frame rate of an animation
   • The compression ratio of a texture
7. What does the original NeRF network predict from a 3D location and viewing direction?
   • A binary object label and no appearance information
   • A scene’s legal ownership and camera price
   • Volume density and view-dependent emitted radiance
   • A single pixel mask with no view dependence
8. What scene representation does 3D Gaussian Splatting optimize for real-time radiance-field rendering?
   • A single triangle silhouette for the entire scene
   • A text-only prompt with no geometric representation
   • A depth map with no color or camera information
   • A set of 3D Gaussians with position, covariance, opacity, and appearance parameters