16-889 Assignment 1: Rendering Basics with PyTorch3D

Bowei Chen boweiche@andrew.cmu.edu

To reproduce the result, please run:

python starter/main.py

1. Practicing with Cameras

1.1. 360-degree Renders (5 points)

The 360 video is shown below:

1.2 Re-creating the Dolly Zoom (10 points)

2. Practicing with Meshes

2.1 Constructing a Tetrahedron (5 points)

The mesh should have 4 vertices and 4 faces.

2.2 Constructing a Cube (5 points)

The mesh should have 8 vertices and 12 faces.

3. Re-texturing a mesh (10 points)

I set color1 = [0, 0, 1] and color2 = [1, 0, 0]

4. Camera Transformations (20 points)

Top Left: R represents rotating 90 degree along z axis, which is [[0, 1, 0], [-1, 0, 0], [0, 0, 1]]. T is 0.

Top Right: R is identity, and T is [0, 0, 3]

Bottom Left: R is identity, and T is [1,-1,1]

Bottom Right: R represents rotating 90 degree along y axis, which is [[0, 0, 1.0], [0, 1, 0], [-1.0, 0, 0]]. T will translate the object back to the original position before rotation, which is [3, 0, 3]

5. Rendering Generic 3D Representations

5.1 Rendering Point Clouds from RGB-D Images (10 points)

5.2 Parametric Functions (10 points)

5.3 Implicit Surfaces (15 points)

Point cloud is more memory efficient than mesh because mesh need to store faces.

Generally, mesh can provide better rendering quality since it tends to provide smooth results.

The rendering speed of point cloud is faster than that of mesh.

6. Do Something Fun (10 points)

I replace the texture of the cow with a wooden texture, producing a wooden cow.