16-825 Assignment 1¶
Kaushik Chavali (kchavali)
1. Practicing with Cameras¶
1.1 Cow 360-degree render¶

1.2 Dolly zoom¶

2. Practicing with Meshes¶
2.1 Tetrahedron¶
A tetrahedron is represented with 4 vertices and 4 triangular faces.

2.2 Cube¶
A triangle mesh cube uses 8 vertices and 12 triangular faces.

3. Re-texturing a Mesh¶
Here, color1 is [0,0,1] (blue, front of the cow) and color2 is is [1,0,0] (red, back of the cow).

4. Camera Transformations¶
| Image Number | Rendered image | Explanation |
|---|---|---|
| Transform 1 | ![]() |
R_relative=[[0, 1, 0], [-1, 0, 0], [0, 0, 1]], T_relative=[0, 0, 0]. Rotates the camera 90° around z axis so the cow appears rotated in the image; no extra translation. |
| Transform 2 | ![]() |
R_relative=[[1, 0, 0], [0, 1, 0], [0, 0, 1]], T_relative=[0, 0, 2]. Leaves orientation alone and moves the camera farther along z by 2 units so the cow looks smaller. |
| Transform 3 | ![]() |
R_relative=[[1, 0, 0], [0, 1, 0], [0, 0, 1]], T_relative=[0.5, -0.5, 0]. Leaves orientation alone and shifts the camera in x/y so the cow is offset in the frame. |
| Transform 4 | ![]() |
R_relative=[[0, 0, 1], [0, 1, 0], [-1, 0, 0]], T_relative=[-3, 0, 3]. Rotates the camera about y axis to a side view and adds translation so the cow stays in view from that new pose. |
5. Rendering Generic 3D Representations¶
5.1 Point clouds from RGB-D images¶
| First RGB-D view | Second RGB-D view | Union of both views |
|---|---|---|
![]() |
![]() |
![]() |
5.2 Parametric functions¶
Parametric torus¶
For $R=3$, $r=1$, and $\phi,\theta\in[0,2\pi]$, the torus is parameterized by
$$ \begin{aligned} x &= (R+r\sin\theta)\cos\phi,\\ y &= (R+r\sin\theta)\sin\phi,\\ z &= r\cos\theta. \end{aligned} $$

Parametric Möbius strip¶
For $u\in[0,2\pi]$ and $v\in[-1,1]$, the Möbius strip is parameterized by
$$ \begin{aligned} x &= (2+0.5v\cos(u/2))\cos u,\\ y &= (2+0.5v\cos(u/2))\sin u,\\ z &= 0.5v\sin(u/2). \end{aligned} $$

5.3 Implicit surfaces¶
Implicit torus¶
For $R=3$, $r=1$, the torus is the zero level set $F(x,y,z)=0$ of
$$ F(x,y,z)=(\sqrt{x^2+y^2}-R)^2+z^2-r^2. $$

Implicit Steiner / Roman surface¶
The Steiner surface is the zero level set $F(x,y,z)=0$ of
$$ F(x,y,z)=x^2y^2+y^2z^2+z^2x^2-xyz. $$

Mesh versus Point Cloud¶
- Quality: Meshes give continuous surfaces with proper shading and silhouettes. Point clouds look sparse or holey unless you use many points, and they don’t fill faces between samples.
- Speed: Point clouds are often faster to rasterize, especially at low sample counts. Dense meshes (many vertices and faces) can be slower to rasterize (like in Q6).
- Memory: A mesh stores vertices + faces (extra connectivity information). A point cloud stores only point locations (sometimes color).
- Ease of use: Parametric surfaces and RGB-D data can be converted into point clouds easily through sampling. Implicit surfaces require running marching cubes to be converted to meshes. Sampling a mesh → point cloud (like in Q7) is easier compared to going the other way around.
- Conclusion: Use meshes when you care about solid appearance and shading. Use point clouds when you have raw samples (depth, parametric surfaces).
6. Extra Credit: Do Something Fun¶
- Loaded a Black Russian Terrier mesh (data/13459_Black_Russian_Terrier_v1_L3.obj) from Free3D.
- Retextured with per-vertex surface normals: maping n↦(n+1)/2 so XYZ directions become RGB.
- Rendered a 360° turntable gif with dist=100, elevation 45°, and a point light at [0, 0, -50].

7. Extra Credit: Sampling Points on Meshes¶
I sampled points uniformly over the cow surface by selecting triangles in proportion to their areas and then sampled barycentric coordinates within each selected triangle.
| Mesh | 10 points | 100 points | 1000 points | 10000 points |
|---|---|---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |










