Curly-Cue: Geometric Methods for Highly Coiled Hair | Hacker News Reader

Curly-Cue: Geometric Methods for Highly Coiled Hair

cs.yale.edu

184 pointscainxinth29 comments

This presentation by one of the authors is a good primer on the current issues around Black hair and hairstyles in Hollywood, CGI, and video games, and why this research is important: https://www.tkim.graphics/MORETHAN/Darke_Slides.pdf

The core issue:

> There have been no papers on afro-textured hair at SIGGRAPH ever.

There are over 300 SIGGRAPH papers on other types of hair.

More accurately, there are over 300 SIGGRAPH papers on hair.

Slide 3, with this headline:

> Curly hair in graphics research is limited to “classical European locks”

Contains this phrase:

> which varies from an elongated ellipse for African hair

Which leads me to think the Eurocentricity of the other 300 papers is being rather oversold. The Beethoven-looking redhead on the previous slide has two hair styles which are unmistakably models of common African hair textures. Being generous I would guess that the hair was made lighter to show details better? Caucasians presenting with that level of frizz is also not unheard of.

Afro-textured hair is certainly a topic worth a few SIGGRAPH monographs all on its own, and a dedicated library is also great. But I suspect that a thorough review of the 300 papers in question would find that the insinuation that none of them treat on the subject of African hair is somewhere between "not supported" and "wildly false".

I'm inclined to believe that the authors, if they quoted that, read it. The implied claim is that this treatment is inadequate. (My reading is supported by the text on the previous slide: page 23 in the PDF.)

I am not going to read through all 328 papers to fact check this, but I'm very confident they did do a "thorough review", that's how you start research like this. There's a bit more detail in the course that this presentation is from:

> As of 2021, we only found two technical papers that showed Afro-textured hair [Bertails et al. 2005; Patrick et al . 2004]. While some technical papers containing Black hair have started to appear since we published our findings [Hsu et al . 2023; Wang et al. 2023], these usually represent relatively unstyled hair. Some styles have started to appear in short talks [Ogunseitan 2022], but a wide range of intricate, sophisticated, and very common hair styles still lie outside the visual language of computer graphics research.

- https://dl.acm.org/doi/10.1145/3664475.3664535 (not open access, I can share it if you're interested though)

The four papers they mention here are not from SIGGRAPH. The "short talk" was from SIGGRAPH 2022, but as that phrase implies it was a talk and not a paper.

One could read the claim being made as there aren’t any papers that are primarily about Afro-textured hair, nor about Black hair geometry considerations, which is likely true. There might not be that many papers that you could say are primarily about Caucasian hair either, but there most definitely is a lot of straight and wavy hair. I think some Siggraph hair papers do show examples of Afro-textured hair, IMO Darke’s slides here have one (p.23, example (c)) tkim.graphics/MORETHAN/Darke_Slides.pdf

Might be worth pointing out there aren’t that many siggraph papers on procedural hair geometry generation, quite a few are on rendering hair and do cover dark/black hair with elliptical cross sections. IIRC there are some papers on blonde hair specifically because it exhibits more visible scattering.

Anyway… is there a historical/cultural bias toward Eurocentric hair and not a lot of representation of Black hair? Yes that would probably be totally fair to say. Splitting hairs on whether or not their words are unambiguously and perfectly accurate might be slightly beside the point. Let’s take it as a hint that we can improve, and enjoy a paper that brings new hair generation methods to bear.

Could a non-black person even present a paper on afro-textured hair, without being subject to the vicissitudes of the social-justice-twitter-sphere?

Yes they could. The paper we're discussing was written by non black people. Another recent paper they reference was as well: https://research.nvidia.com/labs/toronto-ai/adaptive-shells/.

The paper discussed has A.M. Darke listed as an author, who appears to be black. And that adaptive shell paper is not about black hair - it merely uses black hair as an example, along with cat hair, dog hair, leaves, and horse hair.

Ok I don't know what to tell you then, we can't answer your question because there are no papers that meet your criteria. That's the whole point of this discussion, nobody has even tried.

I'm assuming you won't think this counts either, but here's another paper they reference that includes afros, again written by non black authors: https://graphics.cs.utah.edu/research/projects/sag-free-hair....

Black person here; this social disconnect is of course something I pay quite a bit of attention to.

The answer, typically, is OF COURSE. Big time. It would be well appreciated by most.

That being said, I get where this perception comes from -- there can be, as you call them, outsider social-justice folk who might try to say something weird here:

But I just need to highlight the "outsider" part of that -- if only for non-black folks to understand and to try to pay attention to exactly who you're listening to and getting authority from? I hate to use a phrase like "real Black people" but I'll go with that for now; real Black people tend to be the MOST reasonable, but often under-heard.

I'm reminded of, e.g. police reform; I understand that it's incredibly important and necessary. Which is why phrases/ideas like "Abolish the police" and "ACAB" are deeply unhelpful.

(and as always, I am only one Black person, nothing I say here should be taken as gospel for everyone, I could be wrong)

I'm reminded of college in Hyde Park: the older black people wanted the UCPD in the neighborhood because they saw it as a force of stability, while young white students who wouldn't be affected by shrinking the patrol area criticized it.

This leaves out the perspective of the young black people who would have been the disproportionate target of police interactions.

Unfortunately, moderation within a biased status quo doesn't achieve the goal of eliminating the bias. Those seeking a radical change can't reasonably be faulted for using radical rhetoric, only for wanting to change something that's "working."

I usually approach “inclusivity”-driven work with a massive amount of skepticism, but as a black man with highly coiled hair, I have to admit I’m very impressed by this research and work.

I used to be really into gaming when I was younger, and a somewhat disappointing aspect of many games for me was character selection. Even when there were dark-skinned characters, I can’t think of a single game within which I could select or build a character that truly looked like me (or close enough), and usually, it came down to the hair. At the end of the day, it was never a big deal and I still had lots of fun, but it would’ve certainly been nice to have better, more relatable, selections, so I appreciate this research!

The short hair example in particular is remarkably realistic. Immediately familiar to this guy who spent most of his childhood with a skin fade, and a massive improvement over the flat image texture or bump maps that are usually used.

Because many black people (especially African-Americans) have a mixture of hairtypes, I expect the results to be even better once something like this is implemented alongside existing systems.

As someone who’s spent a lot of their career working on digital humans and respective tooling, seeing research into curly hair is a breath of fresh hair.

It’s really difficult to approximate the same way that straight hair is done and there is definitely a very high amount of Eurocentric R&D in this space, which means anything other than a slight frizz doesn’t move right, and has difficulty in rendering correctly. Hair weight specifically is one that is extremely hard to get right and even harder when it’s wet.

I also really recommend watching Theodore Kims talks. He’s ex Pixar and highly respected in the field. He’s done fantastic work in highlighting the lack of research for non-straight hair and non-white skin among other things. This talk has been hugely impactful on me and my own work that many millions of people have seen on the big screen and not realized were digital, in no small part because it highlighted what we were missing.

https://youtu.be/ROuE8xYLpX8?si=qUJ0BvcPDyEKOYLk

People might not understand the importance, but it’s hugely significant to call out the lack of research so that we as an industry can fill that hole and improve our own work. We need to represent a wide range of digital humans , and a lack of technology/research to represent them means our work doesn’t hold up as strongly.

How much processing power does this require? Can it be done in real time?

It probably depends on what you mean by real-time. Hair isn't something I deal with a lot, but I'm familiar with the space.

It says the top model had a few hundred thousand hairs. Doing real-time flat-shaded grooming in a DCC that supported this type of generation is likely doable. However, I haven't seen any real-time hair simulation that wasn't clever shader manipulation in a game engine or the like. It's just too much geometry (including constraints and self-collisions) for the hardware we have now while simulating everything else. And then you've got to shade all of those individual hairs which wuld have a ton of surface area. It would be super resource intensive.

This is more for movie VFX where rendering times are measured in hours per frame on large rendering farms.

This is for hair curve generation, so it can be done offline. Animation and rendering on the resulting models can still be done in real time, to the same degree it could before.

I wonder if this becomes relevant also in analyzing real hair, for designing hair-related products etc.

There is a trillion dollar market in hair, but it is not about virtual models.

I once heard Pat Hanrahan (who among other things was involved in some of the early realistic hair rendering work) claim that L'Oréal funded some of that work, exactly because they were interested in what the models said about the (control of) appearance of real hair.

interesting, why is that? What is that market more into?

I’d never rule out crazy applications of math, but the problems seem pretty orthogonal. When you’re rendering, you want to stipulate by fiat that the hair is perfectly healthy and immune to damage and holds its style no matter what the environment brings to bear, and those are exactly the things most hair care products are meant to help with.

Are they just afraid to use the word "afro"?

None of the hairstyles depicted in the source link are afros, although I understand why the difference could be confusing if you’re not familiar with type 4 hair. An afro is a specific style where the hair is combed out into a spherical puffball.

This looks great, but 234 Kelvin is almost -40° Celsius. Why so cold?

In the case that people don't understand your comments: uppercase K is Kelvin, kilo should be lowercase k. Plenty of people get it wrong, even researchers and scientists who should know better.

Reply on news.ycombinator.com