16-825 Assignment 1: Rendering Basics with PyTorch3D

Andrew ID: tzuyiy

1.1 360-degree Renders

cow turntable

1.2 Re-creating the Dolly Zoom

dolly zoom

Camera distance adjusts with FOV so the cow's size stays constant.

2. Practicing with Meshes

2.1 Tetrahedron

tetrahedron

4 vertices, 4 triangle faces.

2.2 Cube

cube

8 vertices, 12 triangle faces.

3. Re-texturing the Cow

retextured cow

color1 = [0,0,1] (blue, front), color2 = [1,0,0] (red, back), interpolated by z.

4. Camera Transformations

transform 1

1: 90° roll about the Z axis.
R_relative=[[0,1,0],[-1,0,0],[0,0,1]]
T_relative=[0,0,0]

transform 2

2: 90° yaw about the Y axis.
R_relative=[[0,0,1],[0,1,0],[-1,0,0]]
T_relative=[-3,0,3]

transform 3

3: Translation along Z.
R_relative=I
T_relative=[0,0,2]

transform 4

4: Translation along X/Y.
R_relative=I
T_relative=[0.45,-0.5,0]

5. Rendering Generic 3D Representations

5.1 Point Clouds from RGB-D Images

plant point clouds

Left to right: point cloud 1, point cloud 2, and their union (fills in gaps).

5.2 Parametric Functions

torus point cloud

Torus point cloud from its parametric equation. Hole visible from top.

peanut point cloud

Bonus object: a "peanut" shape (two merged spheres), also sampled parametrically.

5.3 Implicit Surfaces

torus mesh

Torus mesh via an implicit function + marching cubes. Hole visible from top.

peanut mesh

Same peanut shape as 5.2, this time built implicitly (union of two spheres), for direct comparison.

Mesh vs. Point Cloud

Point CloudMesh
SpeedFasterSlower with more faces
QualityGaps if sparseSmooth surface
ConstructionJust sample pointsNeeds marching cubes / topology
MemoryPoints + colors onlyVertices + face connectivity
Best forRaw sensor dataPrecise geometry, rendering quality

Extra Credit

6. Do Something Fun

checkerboard cow

Generated a 2D black/white checkerboard image, then used the cow's UV coordinates to map each point on that 2D image onto its corresponding point on the 3D surface (TexturesUV), wrapping the flat pattern onto the mesh.

7. Sampling Points on Meshes

mesh sampling comparison

Left to right: original mesh, then 10 / 100 / 1000 / 10000 sampled points.