16-889 Assignment 1: Rendering Basics with PyTorch3D

1. Practicing with Cameras

1.1. 360-degree Renders (5 points)

To run: python3 main.py --question 1.1. The output will be saved as gifs/cow.gif.

1.2 Re-creating the Dolly Zoom (10 points)

To run: python3 main.py --question 1.2. The output will be saved as gifs/dolly.gif.

2. Practicing with Meshes

2.1 Constructing a Tetrahedron (5 points)

To run: python3 main.py --question 2.1. The output will be saved as gifs/tetrahedron.gif.

The tetrahedron mesh is composed of 4 vertices, and 4 faces.

2.2 Constructing a Cube (5 points)

To run: python3 main.py --question 2.2. The output will be saved as gifs/cube.gif.

The cube mesh is composed of 8 vertices, and 12 triangular faces.

3. Re-texturing a mesh (10 points)

To run: python3 main.py --question 3. The output will be saved as gifs/textured_cow.gif.

The above textured cow is using colors: [0, 0, 1] (Blue) and [1, 0, 0] (Red).

4. Camera Transformations (20 points)

To run: python3 main.py --question [4.1,4.2,4.3,4.4]. The output will be saved under images/.

Question	Image	Rotation	Translation
4.1		Along Z-axis left 90 degrees	None
4.2		None	Along Z-axis by 2 meters
4.3		None	Along X and Y axes by 0.5 meters each
4.4		Along Y axis by 90 degrees	-3 meters in X axis, and 3 meters in Y axis

5. Rendering Generic 3D Representations

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

To run: python3 main.py --question 5.1. The output will be saved as gifs/pointcloud1.gif, gifs/pointcloud2.gif, and gifs/pointcloud3.gif.

Pointcloud 1	Pointcloud 2	Pointcloud 3

5.2 Parametric Functions (10 points)

To run: python3 main.py --question 5.2. The output will be saved as gifs/torus_100.gif, gifs/torus_1000.gif.

Torus w/ 100	Torus w/ 1000

5.3 Implicit Surfaces (15 points)

To run: python3 main.py --question 5.3. The output will be saved as gifs/torus_implicit.gif.

Rendering as a mesh is way faster than as a point cloud and is also higher resolution. In addition, the mesh took less memory as it only needs to store the parameters of the function, whereas a point cloud will need to store each and every single point. In terms of ease of use, a point cloud may be easier depending on how complex the structure is.

6. Do Something Fun (10 points)

For this assignment, I have rendered a Pokeball animation.

The Pokeball mesh was obtained from Free3D. However, the mesh did not come with any texture, so the texture is applied manually. This is done by applying multiple layers of coloring starting with red and white for the semi-spheres, and then coloring each inner circle in the middle.

Lastly, we angle the camera and rotate it slowly to give the impression that the pokeball is rotating to show the catching animation. At the end, we reduce the lighting to give the impression that the Pokemon was successfully caught.

(Extra Credit) 7. Sampling Points on Meshes (10 points)

To run: python3 main.py --question 7 --num_samples 10000. The output will be saved as gifs/sampled_10000.gif.

Mesh	10 Points	100 Points	1,000 Points	10,000 Points