ASSIGNMENT-1

Submitted by Ambareesh Revanur (arevanur)

Number of late days used is 0.

Q1

Q1.1 Cow Mesh

Q1.2 Dolly zoomed

Q2

Q2.1 Tetrahedron

Mesh requires 4 vertices and 4 triangles to be specified.

Q2.2 Cube

Mesh requires 8 vertices and 12 triangles to be specified.

Q3

color1 is Blue and color2 is Red. I used the same colors specified in the example.

Q4

I have included the rotation matrices and the associated translation vectors I used in the code submission.

R_rel indicates the Rotation matrix of camera2 relative to camera1 and T_rel indicates the shift imposed on camera1 to yield camera2.

Q5

Q5.1

Q5.2

Q5.3

Tradeoffs b/w point cloud and a mesh

1. Rendering speed: Point cloud is just a projection of 3D points on to a 3D plane. In mesh, the triangles would have to be constructed first and the closed surface inside the triangle is also rendering in mesh. Hence point clouds are faster to render.
2. Rendering quality: The mesh representation is able to render high quality because it models the connectivity information which can be used to texture the 3D surface. However, in point cloud , the points could be colored but they are discrete. Hence, the rendered image looks discontinuous with point clouds.
3. Ease of use: Point cloud is much easier to deal with and can be used to define loss functions for easy training. Mesh is sensitive to permutations and ordering of vertices need to respected.
4. Memory: Point cloud wins over meshes, since meshes additionally needs to keep track of connectivity information.

Q6

Colosseum Fly-through

Key Idea: Decelerate camera as you move in and accelerate the camera as you move out.

Q7

10 points

100 points

1000 points

10000 points