Lights and Shadows (2020)

The best programmers are cautious about dependencies. Taking something is easy. But you don't learn. And you give up control.

Honestly, if you’re wanting to produce something as good as this, Three.js or other such things just aren’t particularly helpful. It is easier to just ignore all the libraries and do it all from scratch. Popular libraries are good at producing finished products in a particular shape. When you’re wanting to demonstrate the implementation steps and allow intricate fiddling and have everything polished like you want it, they’re generally somewhere between painful and hopeless.

You could still keep Three.js for bits like vector calculations, but it just doesn’t feel worth it, it’s easy enough to implement yourself—or copy and paste from some such library and modify as needed—and will be much lighter. And you build up the bits and pieces you need over time.

What are some good sources to learn this kind of graphics programming work?

In your first link the narrator says he "doesn’t understand the physics of it" but there's really no physics involved (ignoring scatter). It’s just a consequence of the math. It’s relatively easy to understand if you think of it in terms of the surface of a sphere. There is a fixed amount of light coming from a point source, and as the light travels outward you can think of it as being spread over the surface of a sphere. Since the surface area of a sphere is 4pir^2, if you double the radius the area quadruples, and therefore the light intensity at any point on the sphere drops by a factor of four.

edit: And now after rtfm I see there's a nice demo of this!

This is edutainment. Kind of like YouTube channels like 3Blue1Brown, Veritasium or MinutePhysics. All good, it helps build intuition, and to better understand the world, but it is a bit lacking for actually using that knowledge.

Notice how few equations there are in this page, it is a common feature of edutainment, they won't give you equations unless they can't get away without. No linear algebra here, just a cosine (actually a dot product in disguise) and the inverse square law (1/r²), two equations he considered too fundamental to skip. Also notable is the lack of exercises.

But now that you have read this page, and played with the interactive elements, you probably have a good understanding of how lights and shadows work, but can you write a 3D engine or even just calculate exposure time without your camera helping you? Without prior knowledge, probably not. For that, you actually need to do the maths, with exercises and all that. And by the way, look at the source code (it is not obfuscated), all the linear algebra that is not present in the article is there!

That, I think, explain the discrepancy between edutainment and textbooks. Textbooks are for you to do actual work, do the maths, solve problems, etc... not just give you an overview. That's also why is takes way longer and requires a lot more effort on the student part.

Interactive content like this one is good, and maybe it should be given a bit more consideration by the traditional educational system. But I don't think it can replace textbooks, at least not for the "hard part".

It seems most digital attempts aren't done properly, thus it tends to end back to books.

A recent example,

https://www.france24.com/en/tv-shows/focus/20260106-back-to-...

There is the “tools for thought” space I came across where some work felt compelling for computer enabled teaching and learning.

Bret Victor [0], Andy Matuschak [1], and Seymour Papert [2] to point to a few names.

[0]https://www.youtube.com/watch?v=oUaOucZRlmE

[1]https://quantum.country/

[2]https://www.google.com/books/edition/Mindstorms/nDjRDwAAQBAJ...

School incentives are not really aligned around maximizing learning rate for every student. (E.g. that is why there is/was debate around teaching phonetics)

Maybe someone with first-hand experience can weight in, but isn't this what alternative education platforms like "Brilliant" look like?

Yeah I got real excited a new one dropped

One term for them is “explorable explanations”

https://en.wikipedia.org/wiki/Explorable_explanation

https://explorabl.es/

https://samwho.dev/ (e.g. Reservoir sampling, Load balancing)

https://imadr.me/pbr/ (physically based rendering)

You can start by looking at all his other entries: https://news.ycombinator.com/from?site=ciechanow.ski

There is a solution called Radiance Cascades [1] which doesn't require a denoiser for rendering real-time shadows for volumetric lights. Unfortunately the approach is relatively slow, so solutions based on denoising are still more efficient (though also expensive) in terms of the quality/performance tradeoff.

One issue with modern ReSTIR path tracing is that currently the algorithm relies on white (random) noise, which contains low-frequency (large-scale) noise, which produces blotchy boiling artifacts at low sample counts. Optimally an algorithm should use some form of spatio-temporal blue noise with exponential decay to only get evenly distributed high frequency samples. But that's still an open research problem.

1: https://youtube.com/watch?v=rG2aok2SdbU

Having come from graphics in the 90's, practical high-performance answers typically involve fakery on both primary surface shading and shadow calculation.

I've pulled some tricks like "object-pre-pufficiation" (low-frequency model manifold encapsulation, then following the same bones for deformation) mixed with normal recording in shadow layers (for realtime work on old mobile hardware), but, these days, so much can be done with sampling and proper ray-tracing, the old tricks are more novelty than necessary.

It still pays to fake it, though.

bb wake up, new chichanow.ski just dropped

oh wait, (2020) :(

thankfully i don't remember much from this one, so was able to extract some dopamine from it still