16-889: Learning for 3D Vision (SP22) - Assignment 1

• Name: Yen-Chi Cheng (yenchich@andrew.cmu.edu)
• AndrewID: yenchich

Q1

• 1.1:

• 1.2:

Q2

• 2.1
  • Number of vertices: 4. Number of faces: 4.

• 2.2:
  • Number of vertices: 8. Number of faces: 12.

Q3

• color1=[0.3, 0.5, 0.7], color2=[0.7, 0.5, 0.3].

Q4

• 4.1:
  R =

  0	1	0
  -1	0	0
  0	0	1

  
  T =

  0

  0

  0

• 4.2:
  R =

  1	0	0
  0	1	0
  0	0	1

  
  T =

  0

  0

  2

• 4.3:
  R =

  1	0	0
  0	1	0
  0	0	1

  
  T =

  0.5

  -0.5

  0

• 4.4:
  R =

  0	0	1
  0	1	0
  -1	0	0

  
  T =

  -3

  0

  3

Q5

• 5.1

• 5.2

• 5.3

• Mesh vs Point cloud:
  • Rendering speed: slower for mesh while faster for point cloud.
  • Rendering quality: better for mesh compared to point cloud because mesh have faces which make the rendering results more smooth.
  • Ease of use: for training a neural network, point cloud seems to be easier to use than mesh. Because we can treat point cloud as a [N x 3] tensor and directly working with them using neural networks (e.g., some MLPs). While for mesh we need to consider their faces, directly using MLPs on them might not be as effective.
  • Expressiveness: without normals, point cloud loses the information about the geometry compared to meshes.

Q6

Q7