Assignment 0

16-825 Learning for 3D Vision

Name
Harshit Jain
Andrew ID
hjain2
Date
September 1, 2026

A warm-up for the course submission process. I asked an AI assistant how to build a webpage from scratch, then used what it told me to put this page together, including the two figures below.

Task 1. Asking an AI how to build this page

Since the assignment is about learning how webpages are made, I asked an AI assistant (Claude) for a walkthrough rather than for something decorative. The reply below is what I actually built this page from.

Prompt

How do I make a webpage? Full guidance, please.

Response

A webpage is a plain text file that a browser knows how to draw. There is no compiler and no build step needed to get started. Three languages do the work: HTML describes the content and its structure, CSS controls how that content looks, and JavaScript adds behaviour. You can go a long way with only the first two.

Start by creating a file called index.html. Browsers open that filename automatically when someone visits a folder, which is why it is almost always the name of the main page. Put this inside it:

<!DOCTYPE html>
<html lang="en">
<head>
  <meta charset="utf-8">
  <meta name="viewport" content="width=device-width, initial-scale=1">
  <title>My page</title>
  <link rel="stylesheet" href="style.css">
</head>
<body>
  <h1>A heading</h1>
  <p>A paragraph of text.</p>
  <img src="media/photo.jpg" alt="What the photo shows">
</body>
</html>

Double-click that file and it opens in your browser. That is the whole loop: edit, save, refresh. The head holds information about the page, such as its title and which stylesheet to load. The body holds everything a visitor actually sees. Tags come in pairs that wrap their content, so a heading is written as <h1>text</h1>. Images are the exception, since they have nothing to wrap.

Use tags that match what the content means rather than how you want it to look. Headings run from h1 down to h6 and should nest in order. Paragraphs go in p, lists in ul or ol, and a figure with a caption in figure and figcaption. This matters for screen readers, and it gives you sensible things to attach styling to later. Always write an alt description on images so the page still makes sense to someone who cannot see them.

Styling lives in a separate file, style.css, linked from the head. A rule picks elements with a selector and then sets properties on them, so body { font-size: 18px; } changes the text size everywhere. Give an element a class when you want to style one group specifically: <div class="card"> is targeted by .card { ... }. Keeping CSS out of the HTML means you can restyle the whole site by editing one file.

Two things reliably break pages when they move to a server. The first is paths. Writing src="C:/Users/me/photo.jpg" works on your laptop and nowhere else, so always use a path relative to the HTML file, like media/photo.jpg. The second is case: most servers treat Photo.JPG and photo.jpg as different files even though Windows and macOS usually do not. Match the filename exactly.

To publish, copy the folder to whatever web host you have been given and visit the URL. Test it there rather than trusting the local version, because a missing file often only shows up once the page is served. If an image does not appear, open the browser's developer tools and look at the network tab, which will tell you the exact path it tried to load and failed to find.

What stood out

The point about relative paths turned out to be the one that mattered most here, since the course explicitly warns about it and the page has to work from the class server rather than from my laptop. I also liked that the advice was to test on the server instead of locally, which is exactly the failure I would otherwise have discovered after submitting.

Task 2. A trefoil knot mesh

Rather than downloading a picture, I generated one, since making the geometry myself seemed closer to the spirit of the course. The shape below is a trefoil knot: a closed curve that cannot be untangled into a circle without cutting it, which makes it a nice example of something whose interesting property is topological rather than visual.

The surface is a tube swept along that curve. A circle of 16 points is carried around a path of 140 steps and the neighbouring rings are joined into triangles, giving 2,240 vertices and 4,480 faces. Each triangle is shaded flat by the angle between its normal and a single light, so the facets stay visible and you can read the mesh structure off the surface. Getting the ring to line up with itself after a full loop is the fiddly part; using the curve's analytic derivatives gives a frame that closes exactly, where carrying one along numerically leaves a seam.

A trefoil knot rendered as a blue triangle mesh on a dark background, with its wireframe visible over the shaded surface
Trefoil knot as a triangle mesh, 4,480 faces, flat shaded with the wireframe drawn over it. Source: generated by me for this assignment with numpy and Pillow. The script is included in this folder as generate_media.py.

Task 3. Subdivision turntable

For the GIF I made a turntable, which is the format most of the renders in this course seem to use. It starts as a bare icosahedron with 20 faces and refines three times while it spins. Each pass splits every triangle into four and pushes the new vertices out onto the unit sphere, so the face count runs 20, 80, 320, 1,280 and the silhouette tightens onto a circle as the flat faces stop cutting the corners off.

The jump in smoothness between the first two steps is much larger than between the last two, which is the part I find funny about it: the mesh looks basically finished long before the triangle count stops quadrupling.

Animated GIF of an icosphere rotating while its subdivision level increases from 20 faces to 1280 faces
Icosphere turntable cycling through subdivision levels 0 to 3, 32 frames. Source: generated by me for this assignment with numpy and Pillow, using the same generate_media.py script.