Rendering Basics with PyTorch3D¶

16-825 Learning for 3D: Assignment 1

Author: Benny Zhu (haozhenz)
Date: September 2026

Table of Contents¶

  • 1. Practicing with Cameras (15 Points)
    • 1.1 360-degree Renders (5 points)
    • 1.2 Re-creating the Dolly Zoom (10 points)
  • 2. Practicing with Meshes (10 Points)
    • 2.1 Constructing a Tetrahedron (5 points)
    • 2.2 Constructing a Cube (5 points)
  • 3. Re-texturing a Mesh (10 points)
  • 4. Camera Transformations (10 points)
  • 5. Rendering Generic 3D Representations (45 Points)
    • 5.1 Rendering Point Clouds from RGB-D Images (10 points)
    • 5.2 Parametric Functions (10 + 5 points)
    • 5.3 Implicit Surfaces (15 + 5 points)
  • (Extra Credit) 6. Do Something Fun (10 points)
  • (Extra Credit) 7. Sampling Points on Meshes (10 points)

1. Practicing with Cameras (15 Points)¶

1.1 360-degree Renders (5 points)¶

python -m starter.Q1_1

360-degree render of cow

1.2 Re-creating the Dolly Zoom (10 points)¶

python -m starter.Q1_2

Dolly Zoom


2. Practicing with Meshes (10 Points)¶

2.1 Constructing a Tetrahedron (5 points)¶

python -m starter.Q2_1

Tetrahedron

The tetrahedron mesh has 4 vertices and 4 faces.

2.2 Constructing a Cube (5 points)¶

python -m starter.Q2_2

Cube

The cube mesh has 8 vertices and 12 faces.


3. Re-texturing a Mesh (10 points)¶

python -m starter.Q3

Retexture

In this case, color1 = [0.7, 0.7, 0] and color2 = [0, 0, 0.7].


4. Camera Transformations (10 points)¶

python -m starter.Q4 
python -m starter.Q4 --transform_id 1 --output_path images/textured_cow_1.jpg
python -m starter.Q4 --transform_id 2 --output_path images/textured_cow_2.jpg
python -m starter.Q4 --transform_id 3 --output_path images/textured_cow_3.jpg
python -m starter.Q4 --transform_id 4 --output_path images/textured_cow_4.jpg

Original render:

Original


T1: The relative rotation $R_1$ rotates the camera 90° about its Z-axis (the optical axis) with no translation. This causes the cow to appear rotated 90° within the image plane.

$$ R_1 = \begin{bmatrix} 0 & 1 & 0 \\ -1 & 0 & 0 \\ 0 & 0 & 1 \end{bmatrix}, \quad T_1 = \begin{bmatrix} 0 \\ 0 \\ 0 \end{bmatrix} $$

T1


T2: No rotation is applied. The translation $T_2 = [0, 0, 3]$ moves the camera 3 units further along the Z-axis, making the cow appear smaller (further away).

$$ R_2 = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}, \quad T_2 = \begin{bmatrix} 0 \\ 0 \\ 3 \end{bmatrix} $$

T2


T3: No rotation is applied. The translation $T_3 = [0.5, -0.5, 0]$ shifts the camera in the X-Y plane, which moves the cow toward the bottom-left of the frame.

$$ R_3 = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}, \quad T_3 = \begin{bmatrix} 0.5 \\ -0.5 \\ 0 \end{bmatrix} $$

T3


T4: The relative rotation $R_4$ rotates the camera 90° about the Y-axis, giving us a side view of the cow. The translation $T_4 = [-3, 0, 3]$ repositions the camera so the cow remains within the field of view.

$$ R_4 = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ -1 & 0 & 0 \end{bmatrix}, \quad T_4 = \begin{bmatrix} -3 \\ 0 \\ 3 \end{bmatrix} $$

T4


5. Rendering Generic 3D Representations (45 Points)¶

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

python -m starter.Q5 --render point_cloud
Point Cloud 1 Point Cloud 2 Union
PC1 PC2 Union

5.2 Parametric Functions (10 + 5 points)¶

python -m starter.render_generic --render parametric --num_samples 1000
python -m starter.render_generic --render mobius --num_samples 1000
Torus Möbius Strip
Torus Mobius Strip

5.3 Implicit Surfaces (15 + 5 points)¶

python -m starter.Q5 --render implicit
python -m starter.Q5 --render heart
Implicit Torus Implicit Ellipsoid
Implicit Torus Implicit Ellipsoid

Mesh vs. Point Cloud Tradeoffs¶

Mesh Point Cloud
Rendering Speed Slow Fast
Rendering Quality High Depend on the Density of the points
Ease of Use Hard Easy
Memory Usage High Low

(Extra Credit) 6. Do Something Fun (10 points)¶

python -m starter.Q6

Fun Result

Render a dragon. The dataset was downloaded from Free3D, but unfortunately, it does not provide any information about the dataset.


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

python -m starter.Q7 --num_samples 10
python -m starter.Q7 --num_samples 100
python -m starter.Q7 --num_samples 1000
python -m starter.Q7 --num_samples 10000
Original Mesh 10 points 100 points 1,000 points 10,000 points
Mesh 10 100 1000 10000