Plenoxels and Voxel Radiance Fields Quiz

Answers and explanations appear as you take the quiz.

1. What is the most surprising design choice in Plenoxels compared to NeRF?
   • It uses no neural network at all
   • It uses a much deeper neural network
   • It only works in 2D
   • It requires LiDAR scanners
2. What does each occupied voxel in Plenoxels store to capture view-dependent color?
   • A full RGB texture map
   • Spherical harmonic coefficients
   • A separate neural network
   • A depth value only
3. What optimization-time comparison did the Plenoxels paper report on its evaluated benchmarks?
   • Exactly the same optimization time on every scene
   • About two orders of magnitude slower in all cases
   • About two orders of magnitude faster than NeRF in the reported settings
   • No measurable difference under any setup
4. How are values fetched from the voxel grid along a ray?
   • Nearest-neighbor lookup only
   • Random sampling
   • A convolutional filter
   • Trilinear interpolation
5. What conclusion is supported by Plenoxels’ benchmark results?
   • An explicit sparse grid can achieve NeRF-comparable results on the evaluated tasks without an MLP
   • Neural networks are never useful for 3D vision
   • Voxels always outperform every neural representation
   • Volume rendering requires no optimization
6. To stay efficient, how is the Plenoxels voxel grid stored?
   • As a dense fully filled cube
   • As a sparse grid keeping only occupied voxels
   • As a single flat image
   • As compressed video frames
7. What regularizer did the Plenoxels paper use to encourage spatial smoothness?
   • A speech-recognition loss
   • A KL penalty between language models
   • A total-variation regularizer on voxel coefficients
   • A bitrate constraint on rendered images
8. What does “Plenoxels” expand to?
   • Planar neural outputs
   • Perceptual linear voxels
   • Point-level network samples
   • Plenoptic voxels