Point Cloud Processing Quiz

Answers and explanations appear as you take the quiz.

1. What is a point cloud?
   • A collection of 3D points describing surfaces in space
   • A 2D grid of color pixels
   • A cluster of servers processing graphics
   • A weather map of cloud cover
2. Why can't standard image CNNs be applied directly to point clouds?
   • Point clouds are too small
   • Points are unordered and lack a regular grid structure
   • Point clouds contain no spatial information
   • CNNs only work in black and white
3. What key property must a point cloud network respect?
   • Color invariance
   • Strict left-to-right reading order
   • Permutation invariance to point ordering
   • Fixed image resolution
4. How did PointNet achieve order-independence?
   • By sorting points alphabetically
   • By converting points to pixels first
   • By discarding all but the first point
   • By using a symmetric function like max-pooling over per-point features
5. What does the Iterative Closest Point (ICP) algorithm do?
   • Aligns two point clouds into a common coordinate frame
   • Compresses point clouds for storage
   • Colors points by height
   • Detects faces in 3D scans
6. Why is voxel downsampling commonly used?
   • To add more points for detail
   • To reduce huge, dense clouds to a manageable, more uniform size
   • To convert 3D data into audio
   • To encrypt the point cloud
7. What does PointNet++ add beyond PointNet’s independent per-point feature processing?
   • A fixed left-to-right order for every point cloud
   • An image-only convolution applied to RGB pixels
   • Hierarchical grouping of nearby points to learn local features at multiple scales
   • A rule that deletes every point outside the origin
8. Why might a point-cloud pipeline remove statistical outliers before modeling?
   • To create an ordered sequence from the remaining points
   • To convert the cloud into a two-dimensional audio spectrogram
   • To add synthetic surface normals to every point
   • To discard isolated noisy points that do not fit the surrounding point distribution