Assignment 1: Rendering Basics with PyTorch3D¶

Name: Chenzhe Zhang

Course: 16-825 Learning for 3D Vision

1.1 360-degree Renders¶

1.1 360-degree Renders

1.2 Re-creating the Dolly Zoom¶

1.2 Re-creating the Dolly Zoom

2.1 Constructing a Tetrahedron¶

4 vertices and 4 triangular faces

2.1 Constructing a Tetrahedron

2.2 Constructing a Cube¶

8 vertices and 12 triangular faces

2.2 Constructing a Cube

3. Re-texturing a Mesh¶

I chose color1 = [0, 0, 1] (blue) for front and color2 = [1, 0, 0] (red)for back.

Cow with blue-to-red vertex colors

4. Camera Transformations¶

transform1¶

The relative rotation rotates the cow's projection clockwise by 90 degrees, so its head points to the right. The relative translation is zero, leaving the camera distance unchanged.

$$ R_{\mathrm{relative}} = \begin{bmatrix} 0 & 1 & 0 \\ -1 & 0 & 0 \\ 0 & 0 & 1 \end{bmatrix} $$

$$ T_{\mathrm{relative}} = [0,0,0] $$

Cow rotated in the image plane

transform2¶

The relative rotation is the identity matrix, preserving the cow's orientation. The relative translation increases the camera-space depth from 3 to 6, making the cow appear smaller.

$$ R_{\mathrm{relative}} = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} $$

$$ T_{\mathrm{relative}} = [0,0,3] $$

$$ T = [0,0,6] $$

Cow viewed from farther away

transform3¶

The relative rotation is the identity matrix. In PyTorch3D's camera coordinates, positive x moves the projection to the left, and negative y moves it downward. The depth remains unchanged.

$$ R_{\mathrm{relative}} = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} $$

$$ T_{\mathrm{relative}} = [0.5,-0.5,0] $$

$$ T = [0.5,-0.5,3] $$

Cow shifted left and down

transform4¶

The relative rotation rotates about the y-axis to show the cow from the side, with its head on the left. The relative translation compensates for the rotated initial translation and keeps the cow centered.

$$ R_{\mathrm{relative}} = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ -1 & 0 & 0 \end{bmatrix} $$

$$ T_{\mathrm{relative}} = [-3,0,3] $$

The final translation is:

$$ T = R_{\mathrm{relative}}T_0 + T_{\mathrm{relative}} = [3,0,0] + [-3,0,3] = [0,0,3] $$

Side view of the cow

5.1 Rendering Point Clouds from RGB-D Images¶

From left to right: the point cloud from image 1, the point cloud from image 2, and their union.

Point clouds from image 1, image 2, and their union

5.2 Parametric Functions¶

Torus¶

Torus point cloud

Additional Object: Ellipsoid¶

Mobius strip point cloud

5.3 Implicit Surfaces¶

Torus¶

Torus mesh

Mesh vs. Point Cloud¶

Meshes produce smooth, connected surfaces, while sparse point clouds may show gaps. Point clouds are easier to create because they do not need faces. Rendering speed and memory usage depend on the number of points or triangles and the rendering settings.

Additional Object: Ellipsoid¶

Ellipsoid mesh

6. Do Something Fun¶

I designed three color schemes for a sofa, assigning different vertex colors to the cushions, frame, and legs. I rendered a synchronized 360-degree comparison of the three designs.

Three sofa color schemes

Model: Sofa by berkayc on Free3D.

7. Sampling Points on Meshes¶

Sampled cow point clouds and original mesh