ZOZO's Contact Solver for physics-based simulations
github.com53 points by vintagedave 6 hours ago
53 points by vintagedave 6 hours ago
For those that don't know, ZOZO is a tech-forward clothing designer/retailer.
Several years back, they sent me a special spandex shirt/leggings combo, black with spaced white dots. Then you use their app to take many photos of yourself, and they have a profile of your body to be used for automatic fitting.
The shirt they eventually sent did fit well, but wasn't anything special several years ago.
This shows that they are still at it, and as someone that hates shopping for clothing, I hope this is a sign that the dream of a custom tailored fit at a mass production price is getting nearer.
This seems to be the relevant Two Minute Papers with a very quick explainer.
Was Two Minute Papers always so sensationalistic or is that a recent change? I remember seeing the videos many many years ago, and don't recall him being so overly enthusiastic and borderline sensationalistic, like this video seems to be.
Even the title of the video is straight up clickbait ("The Worst Bug In Games Is Now Gone Forever") since the context is all wrong, the metrics on the top left even shows "time/frame: 3.38 min", how could that be useful for games? The problem with physics in games is in real-time simulations, not in cached/animated "physics".
Don't get me wrong, the simulations are impressive, and hopefully will have a big impact on simulation stability for real-time and not, I was just taken aback by the video.
It has always been his style, you can check for yourself by watching some of his early videos. Over the years, he has refined it and fully committed to it.
I usually don't like too much sensationalism, but he gets a pass. That's just his style and I think he does it well without compromising on the information content. He acknowledges that the technique is slow by the way, but that's late in the video.
But I agree that the title is poorly chosen in this case and I think it would be more appropriate for the previous video about a similar paper [1] where the simulation is less accurate, but runs in real-time. It is as if the titles were swapped.
[1] https://www.youtube.com/watch?v=7NF3CdXkm68
Edit: And of course, it is entertainment, what did you expect of a YouTube channel covering state-of-the-art research in less than 10 minutes! If you want to get serious, read the actual paper. Short(ish) YouTube videos is simply not the right format for serious work, sensationalism or not.
> Short(ish) YouTube videos is simply not the right format for serious work, sensationalism or not.
I disagree. For example SIGGRAPH presentation videos manage to be short, informative, and largely non-sensationalist. You can see some of them in this playlist: https://www.youtube.com/playlist?list=PL1PdIP1lGMJJzRFjlDajK...
StiffGIPC presentation makes good contrast here: https://www.youtube.com/watch?v=3TBoTX2vag4
Yeah, love it or hate it, I do think they thread the needle between genuine excitement and overhyping.
The titles and thumbnails are getting clickbaity though.
I used to watch his videos early on, but it's been like this for a few years at least.
Same here, that's why I was kind of surprised. Shame what YouTube forces creators to degrade into, I remember it being super nice being able to see a video about a new SIGGRAPH paper before diving into the details, but these new videos (well, "new" if what you say is true about it being years) I can barely stand because of the change...
FYI you're responding to a 3-days-old account with way too many comments in such a short time frame to be legit. It's most likely a bot.
Lol, thanks I guess, but I'm just bored and have lots of free time :)
Also, based on my first message in this submission, on on earth (like exactly) would an LLM or something else be able to leave a comment like that? Do spambots on the internet have entire backstories now or what?
He has been like that for couple of years at least. I guess it wins clicks in the youtube slot-machine. But I can't stand him either, despite being exact target audience for his videos.
What value do all the emojis provide?
yeah its pretty funny, i wonder if they prompted the llm to put as many emojis in as possible:
<edit> forgot hn doesnt show emojis, so ill just link to the paragraph: https://github.com/st-tech/ppf-contact-solver?tab=readme-ov-...
8 emojis in 2 sentences, lol
They made me stop reading halfway through.
It didn't help that they make meaningless claims like
> Physically Accurate: Our deformable solver is driven by the Finite Element Method.
I don't know or care if they used an LLM to write that readme, but it's hot garbage. A pity because it seems like a decent sim otherwise.
What's wrong with that statement? FEM is a good way to handle deformables, but it isn't the only way, so it a fine statement.
If I'm understanding correctly, the same approach was implemented also in IPC Toolkit here: https://github.com/ipc-sim/ipc-toolkit/pull/148
Fun fact: In Haiti, "zozo" is a slang term for male genitalia.
Not realtime, seconds-minutes per frame.
Is your comment here to refute a claim you saw somewhere, or to simply point this out? I wouldn’t expect this to be real-time, given the complexity, nor do I believe it needs to be in order to be useful.
It is good to point it out it is for offline simulations. There is some related recent work, Offset Geometric Contact that is suitable for interactive use: https://ankachan.github.io/Projects/OGC/index.html
Also assumed by default we were talking about real-time, but then I saw Python/juPyter and a rendered videos, got a bit confused, then came across "46.4s/frame" for one of the examples and finally registered it wasn't about real-time.
I agree it doesn't have to be real-time to be valid, I think my mindset just goes to physics in video games which are usually real-time when I see contact solvers or most other things related to simulations.
Contact is a hard problem to solve and there's some tangential softwares that do it well within the FEA space. I'd be curious to know how this does with materials/geometries of vastly different stiffnessess and if it produces realistic reaction/contact forces (one cheap way to manage contact is to jack up the contact stiffness, which will prevent penetration, but drive some unrealistic forces at those interfaces).
I can't quite figure out how to install and use this. Perhaps it would be useful if I could install it as a python package, by providing a pyproject.toml or something? I ran warmup.py which is creating venvs for me and doing all kinds of things I don't really want, but when activating the environment it still failed on 'from frontend import App', which seems to be commonly used in your examples.
Contributors:
claude 19 commits, +21,000 lines
That's quite a bad faith take when you'd have seen claude is used at the very end after 10 months of another author's work with +62,847 lines.
Holy emoji batman!
Shirt shells? Tree stump solids? Knot rods?
I have no idea what any of those mean.