16-825 Assignment 1: Rendering Basics with PyTorch3D¶

1. Practicing with Cameras¶

1.1. 360-degree Renders¶

360-degree Renders

1.2 Re-creating the Dolly Zoom¶

Dolly Zoom

2. Practicing with Meshes¶

2.1 Constructing a Tetrahedron¶

  • Vertices: 4
  • Faces: 4

tetrahedron

2.2 Constructing a Cube¶

  • Vertices: 8
  • Faces: 12

cube

3. Re-texturing a mesh¶

  • color1: Same blue color as previous task. ([0.7, 0.7, 1])
  • color2: Red. ([1, 0, 0])

re-texturing a mesh

4. Camera Transformations¶

Image1¶

camera transform1

  • R_relative: 90 degree rotation around the z-axis.
  • T_relative: No translation.

Image2¶

camera transform2

  • R_relative: No rotation.
  • T_relative: Translates the cow 2 units along the positive z-axis.

Image3¶

camera transform3

  • R_relative: No rotation.
  • T_relative: Translates the cow 0.5 units along the positive x-axis, and 0.5 units along the negative y-axis.

Image4¶

camera transform4

  • R_relative: 90 degree rotation around the y-axis.
  • T_relative: Translates the cow 3 units along the negative x-axis, and 3 units along the positive z-axis.

5. Rendering Generic 3D Representations¶

5.1 Rendering Point Clouds from RGB-D Images¶

  1. The point cloud corresponding to the first image
No description has been provided for this image
  1. The point cloud corresponding to the second image
No description has been provided for this image
  1. The point cloud formed by the union of the first 2 point clouds.
No description has been provided for this image

5.2 Parametric Functions¶

Render Torus Point Cloud Render Peanut Point Cloud

5.3 Implicit Surfaces¶

Render Torus Mesh

In addition, discuss some of the tradeoffs between rendering as a mesh vs a point cloud. Things to consider might include rendering speed, rendering quality, ease of use, memory usage, etc. (5 points)

  • Rendering speed: Point cloud rendering is fast, and the mesh rendering is slower. The mesh rendering need to consider about the lighting and texture accross the surface.
  • Rendering quality: If the sampling number of point cloud is less, the quality will be low because the point is sparse and contains a lot of holes. The rendering quality of mesh is better because of the continous surface.
  • Ease of use: Point cloud is easy to get from the sensor like lidar, and easy to use because all the point are independent. The mesh is complex to generate and use because of its surface connectivity.
  • Memory usage: The memory usage is linear to the number of point cloud. If many point clouds, then the memory usage is large. More number of triangle in the mesh, more memory usage. What special is that the size of mesh triangle can be large, which needs many point cloud to represent it, uses less memory than point clouds.

Render Peanut Mesh

(Extra Credit) 6. Do Something Fun¶

Sofa with my favorite color

(Extra Credit) 7. Sampling Points on Meshes¶

The original cow mesh/ sample 10 points/ sample 100 points/ sample 1000 points/ sample 10000 points:¶

360-degree Renders Barycentric with 10 sample Barycentric with 100 sample Barycentric with 1000 sample Barycentric with 10000 sample