16-825 Assignment 1: Rendering Basics with PyTorch3D

Name and Andrew ID: Peter Salmon, psalmon

1. Practicing with Cameras

1.1 360-degree Render

A 360-degree turntable render of the cow mesh

1.2 Dolly Zoom

A dolly zoom animation of the cow mesh

2. Practicing with Meshes

2.1 Tetrahedron

4 vertices, 4 triangular faces.

A rotating tetrahedron mesh

2.2 Cube

8 vertices, 12 triangular faces.

A rotating cube mesh

3. Re-texturing a Mesh

color1 = [0, 0, 1] (front, blue) and color2 = [1.0, 0.65, 0.0] (back, orange).

A rotating cow with a front-to-back color gradient

4. Camera Transformations

Transformation 1

Rotate 90 degrees around the camera z-axis; do not translate.

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

Cow rendered with the first relative camera transformation

Transformation 2

Rotate 90 degrees around the y-axis and compensate for the rotated initial translation to keep the cow centered at the same distance.

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

Cow rendered from the side using the second relative camera transformation

Transformation 3

Increase the camera-space z translation from 3 to 5 to make the cow smaller.

R_relative = [[1, 0, 0],
              [0, 1, 0],
              [0, 0, 1]]
T_relative = [0, 0, 2]

Cow rendered farther from the camera

Transformation 4

Translate in x and y to shift the cow left and downward without resizing it.

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

Cow shifted left and downward using the fourth relative camera transformation

5. Rendering Generic 3D Representations

5.1 Point Clouds from RGB-D Images

From left to right: the point cloud from the first RGB-D image, the point cloud from the second RGB-D image, and the union of both point clouds.

Three rotating plant point clouds shown side by side

5.2 Parametric Functions

Torus Point Cloud

A rotating torus represented as a point cloud

Custom Parametric Shape: Cylinder

A rotating cylinder represented as a point cloud

5.3 Implicit Surfaces

Implicit Torus Mesh

A rotating torus mesh extracted from an implicit surface

Mesh and Point-cloud Tradeoffs

A mesh explicitly connects vertices into faces, so it produces a continuous surface with smooth shading and usually looks solid with relatively few vertices. It also requires valid face connectivity and can be more difficult to construct or edit. A point cloud is simpler and can represent sampled geometry directly, but its appearance depends strongly on point density and point radius. Sparse point clouds show holes, while dense point clouds require more memory and rendering work. For this torus, the mesh looks smoother and more solid, while the point cloud is easier to generate directly from the parametric equation.

Custom Implicit Shape: Cylinder

A rotating cylinder mesh extracted from an implicit surface

Extra Credit 7: Sampling Points on Meshes

From left to right: the original cow mesh followed by point clouds containing 10, 100, 1,000, and 10,000 uniformly sampled surface points.

Cow mesh beside point clouds sampled with four different point counts