16-889 Assignment 1: Rendering Basics with PyTorch3D

Jianchun Chen jianchuc@andrw.cmu.edu

---

To reproduce the result, please run

CUDA_VISIBLE_DEVICES={your gpu id} python main.py

1. Practicing with Cameras

1.1. 360-degree Renders (5 points)

1.2 Re-creating the Dolly Zoom (10 points)

2. Practicing with Meshes

2.1 Constructing a Tetrahedron (5 points)

It contains 4 vertices and 4 triangular faces.

2.2 Constructing a Cube (5 points)

It contains 8 vertices and 12 triangular faces from 6 rectangular faces.

3. Re-texturing a mesh (10 points)

The two colors are chosed as red [1,0,0] and green [0,1,0].

4. Camera Transformations (20 points)

R = [[0,-1,0],[1,0,0],[0,0,1]], t=[0,0,0]

R = [[1,0,0],[0,1,0],[0,0,1]], t=[0,0,1]

R = [[1,0,0],[0,1,0],[0,0,1]], t=[0.5,-0.5,0]

R = [[0,0,1],[0,1,0],[-1,0,0]], t=[3,0,3]

5. Rendering Generic 3D Representations

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

The rendered point clouds from the first, second images and the union are shown respectively.

5.2 Parametric Functions (10 points)

5.3 Implicit Surfaces (15 points)

The time and memory comparison between parametric function with 200*200 points and implicit surface with 64*64*3 voxel are shown in the table below.

\	Parametric Functions	Implicit Surface
Time (s)	2.64	4.49
Memory (MB)	1409	1477

Generally, the parametric function is easier to use with faster speed and lower memory use. The quality depends on the resolution of the grid/voxel we use. However, the advantage of implicit surface is that it is a dense representation of a 3D structure.

6. Do Something Fun (10 points)

To reproduce the result, please prepare a sequence of human model from CAPE dataset and two UV texture map. The texture map I used here is generated by Re-Identification Supervised Texture Generation.

The gif below shows a dynamic human gradually changing its cloth texture.

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

The sampled point clouds from the cow mesh with 10,100,1000,10000 points are shown below respectively.