16-825 Learning for 3D Vision

Assignment 1

Andrew ID: pburujwa

Name: Parth Burujwale

Task 1: Practicing with Cameras

1.1 360-degree renders

360-degree gif video of cow mesh

1.2 Re-creating Dolly Zoom

Re-creating Dolly Zoom

Task 2: Practicing with Meshes

2.1 Constructing a Tetrahedron

360-degree gif video of tetrahedron

This tetrahedron has 4 vertices and 4 triangular faces

2.2 Constructing a Cube

360-degree gif video of Cube

This cube has 8 vertices and 12 triangular faces

Task 3: Re-texturing a mesh

360-degree gif video of re-textured cow

The two colors I used are green and yellow

Task 4: Camera Transformations

Transform-none

R_relative=[[1, 0, 0], [0, 1, 0], [0, 0, 1]], T_relative=[0, 0, 0]

Transform-1

R_relative=[[0, 1, 0], [-1, 0, 0], [0, 0, 1]], T_relative=[0, 0, 0]

Transform-2

R_relative=[[0, 0, 1], [0, 1, 0], [-1, 0, 0]], T_relative=[-3, 0, 3]

Transform-3

R_relative=[[1, 0, 0], [0, 1, 0], [0, 0, 1]] T_relative=[0, 0, 3]

Transform-4

R_relative=[[1, 0, 0], [0, 1, 0], [0, 0, 1]], T_relative=[0.5, -0.5, 0]

Task 5: Rendering Generic 3D Representations

5.1 Rendering point clouds from RGB-D images

All three point clouds

5.2 Parametric Functions

Rendering a Torus
Rendering a object of my choice: helix

5.3 Implicit Surfaces

Rendering a Torus defined by implicit function

I found that meshes usually give smoother and more continuous surfaces because their vertices are connected through triangular faces. This generally gives better rendering quality, especially when lighting and shading are used. Point clouds are simpler to work with since they only store points and their features, but a large number of points may be needed to represent a smooth surface properly. If the sampling density is too low or the point radius is too small, gaps can also appear in the rendered point cloud. Meshes require extra information about how the vertices are connected, which can make them slightly more complicated to construct. On the other hand, point clouds can be generated directly from sampled coordinates or RGB D data. In terms of memory, point clouds only store points and their features, while meshes also need to store the faces. However, a dense point cloud may still use more memory because it can require many more points to represent the same surface accurately.

Rendring an Implicit Infinity

(Extra Credit) 6: Do Something Fun

Rendering a building

(Extra Credit) 7: Sampling points on meshes

Sampling points on cow meshes