Assignment 0: Getting Started with Webpage Submission

Task 1. ChatGPT

Prompt

I'm writing my assignment 0 for the course learning for 3D vision. Please answer the following question:

Task 1. ChatGPT
For this task, play around with ChatGPT. Be creative! Now, mention the prompt and the answer given by ChatGPT in your webpage. You can even mention an interesting conversation.

Some of the prompt examples include:

Write a song/poem/story/joke.

Some puzzle or riddle.

Response from ChatGPT (GPT-5 Auto)

Below are prompt/answer pairs and a short conversation for Assignment 0:

---

1) Poem

• Prompt: Write a 6-line poem about point clouds falling like snow and differentiable rendering.
• ChatGPT Answer:

  Points drift down—quiet, slow—
  a snowfall of xyz.
  Normals whisper where to go,
  loss backpropagates, says “this.”
  Light is learned, not merely thrown:
  we render gradients into bliss.

---

2) Micro-story

• Prompt: Write a ~120-word noir micro-story about a voxel detective solving a missing-mesh case; end with a one-line twist.
• ChatGPT Answer:

  They hired me when the model went holey—chunks missing like teeth in a bad city grin. I traced the gaps through warehouse GPUs, following a trail of corrupted OBJ files and cheap reconstruction jobs. Every witness said the same: “Looked solid in the viewer.” Sure. Everything looks solid until you rotate the camera. In a basement full of stale thermal paste, I found the culprit: a greedy simplifier shaving triangles like a bad barber. I restored the edges, stitched the seams, re-weighted the normals. The client cried when the mesh turned watertight. “How much?” she asked. I lit a cigarette, watched the shader bloom. “On the house,” I said.
  Because it was my mesh in the first place.

---

3) Jokes (short)

• Prompt: Tell me three original one-liners about camera calibration, SVD, and RANSAC.
• ChatGPT Answer:
  • My camera asked for couples therapy—turns out its intrinsics and extrinsics just needed better alignment.
  • I tried to impress a matrix with small talk; it replied, “Let’s just decompose the tension with SVD.”
  • Dating tip from RANSAC: reject outliers quickly, commit once you’ve got a strong consensus.

---

4) Riddle

• Prompt: Make a riddle about epipolar geometry and then solve it.
• ChatGPT Answer:

  Riddle: I am a path in one image ruled by a point in another; every truthful match walks me, though I’m drawn by a matrix that never visits either scene. What am I?
  Solution: An epipolar line, generated by the fundamental matrix.

---

5) Puzzle (geometry)

• Prompt: A depth map is $Z(x,y)=ax+by+c$. What 3D surface is it, and what’s a normal vector?
• ChatGPT Answer:

  It’s a plane. In camera coordinates $(x, y, Z)$ with $Z = ax + by + c$, one normal is proportional to:
  $$(-a,\,-b,\,1)$$
  (any nonzero scale works).

---

6) Puzzle (homography)

• Prompt: How many point correspondences are minimally required to estimate a planar homography between two images, and why?
• ChatGPT Answer:

  Four non-collinear correspondences. A homography has 8 DoF ($3\times3$ matrix up to scale). Each match gives two independent equations, so $4\times2=8$ constraints—just sufficient (assuming points aren’t degenerate/collinear).

---

7) Short “interesting conversation” (excerpt)

• Prompt: Explain epipolar geometry progressively.
• ChatGPT Answer (transcript):
  • User: Explain epipolar geometry in one sentence.
  • ChatGPT: It’s the projective relationship that constrains corresponding points in two views to lie on paired epipolar lines defined by the cameras’ relative pose.
  • User: Now explain it like I’m five.
  • ChatGPT: If two cameras look at the same toy, a spot in one picture tells you a line to search on in the other picture. The match has to sit on that line.
  • User: Give me one formula to remember.
  • ChatGPT: $$ \mathbf{x}_2^{\top}\,\mathbf{F}\,\mathbf{x}_1 = 0 $$ where $\mathbf{F}$ is the fundamental matrix and $\mathbf{x}_1,\mathbf{x}_2$ are corresponding homogeneous image points.

Task 2. 3D shape images

Source

Task 3. Funny GIFs

Source