16-825 Assignment 1: Rendering Basics with PyTorch3D
Name: Santiago Arambulo
Andrew ID: sarambul
1. Practicing with Cameras
1.1. 360-degree Renders
1.2. Re-creating the Dolly Zoom
2. Practicing with Meshes
2.1. Constructing a Tetrahedron
My mesh has 4 vertices and 4 faces.
2.2. Constructing a Cube
My mesh has 8 vertices and 12 faces (2 triangles per cube side).
3. Re-texturing a mesh
My choice of colors was color1 = [0.7, 0.7, 1] (purpleish) and color2 = [0.7, 0.2, 0.2] (redish)
4. Camera Transformations
Description of R_relative and T_relative per image:
| Image |
R_relative |
T_relative |
| 1 |
Turn the camera 90° anticlockwise around the z axis |
Nothing |
| 2 |
Nothing |
Move the camera -1 unit along the z axis |
| 3 |
Turn the camera 9° clockwise around the y axis |
Move the camera towards the z axis, and slightly towards the y axis, and the x axis. |
| 4 |
Turn the camera 90° anticlockwise around the y axis |
Move the camera -3 units along the x axis and 3 units along the z axis |
5. Rendering Generic 3D Representations
5.1. Rendering Point Clouds from RGB-D Images
5.2. Parametric Functions
5.3. Implicit Surfaces
Tradeoffs between mesh and point cloud
-
Quality: Meshes look nicer because they show a continuos surface, while points clouds
require the viewer to fill in the gaps to understand what they are seeing.
-
Rendering speed: Meshes are rendered faster than point clouds. The amount of time it took
to render the answer to question 5.3 was around 7 times longer than for other questions.
-
Ease of use: Points clouds require fewer elements than meshes. Working with a mesh demands
processing the faces in addition to the vertices or points. This makes working with point
clouds slightly easier
-
Memory usage: For the same reason as above, the point clouds might take less space in memory.
A mesh needs to store additional information for the faces, which are not required for point clouds.
6. Do Something Fun
"El gato volador"