Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

1000fps image projection on deforming non-rigid surface

k2.t.u-tokyo.ac.jp

Post reply on HN
Print view View original

121–130 of 133 posts

Re: 1000fps image projection on deforming non-rigid surface

#121

Post by leni536 »

phkahler wrote: post #76
jhallenworld wrote: post #49

Earlier quoted context omitted.

It's been done- https://thefutureofthings.com/3029-the-return-of-the-3d-crys... https://www.youtube.com/watch?v=9U80TK6PAVc

The first I heard of it in the 1980s. TI used a laser to project point on a rotating surface. Then someone used a flexible mirror in front of a speaker. The mirror would flex convex/concave changing the apparent distance to a vector display screen. Just saying a modern 1000fps display could do this much better.

That flexing mirror idea is really smart.

Re: 1000fps image projection on deforming non-rigid surface

#122

Post by vsl »

nirvdrum wrote: post #115
oh_sigh wrote: post #100

Earlier quoted context omitted.

Would the Disney version be considered inferior quality to this new research?

The connotative meaning of “poor” isn’t really anything that isn’t the absolute best. Words have meaning beyond the dictionary definition. “Poor” in this case means you think the researchers did a bad job, not that their work was a step on the path to something better. “Earlier”, or even “dated”, would have been a much more charitable depiction. And plenty of people are deriving joy from the application of that techn…

> in this case means you think the researchers did a bad job

It really does not. You’re inventing things to be offended by. It simply means it is low quality version of the same thing, which it very much is.

In retrospect, Ford’s first cars were a poor version of their 2019 ones. Doesn’t in any way diminish the accomplishments of the past.

Re: 1000fps image projection on deforming non-rigid surface

#123

Post by nirvdrum »

vsl wrote: post #122
nirvdrum wrote: post #115

Earlier quoted context omitted.

The connotative meaning of “poor” isn’t really anything that isn’t the absolute best. Words have meaning beyond the dictionary definition. “Poor” in this case means you think the researchers did a bad job, not that their work was a step on the path to something better. “Earlier”, or even “dated”, would have been a much more charitable depiction. And plenty of people are deriving joy from the application of that techn…

> in this case means you think the researchers did a bad job It really does not. You’re inventing things to be offended by. It simply means it is low quality version of the same thing, which it very much is. In retrospect, Ford’s first cars were a poor version of their 2019 ones. Doesn’t in any way diminish the accomplishments of the past.

I’m really not. And I don’t get offended by very much. Maybe you’re not a native speaker? “Poor” as it refers to quality is a negative qualifier. And it’s typically used in a subjective context. That’s the connotative meaning.

Even the definition used here (taken from Google) indicates so, if it weren’t truncated:

poor (adjective) worse than is usual, expected, or desirable; of a low or inferior standard or quality.

The list of synonyms is even more telling about it’s real meaning: shoddy, bad, deficient, defective, lamentable, deplorable, awful, etc.

These are not words I’d use to describe the technology, having seen it first-hand. The technology is not worse than usual and I really don’t see how it’s worse than expected or is otherwise undesirable. It’s out there and being enjoyed by people.

Re: 1000fps image projection on deforming non-rigid surface

#125

Post by rebuilder »

timdorr wrote: post #8

The second generation of the projection tech does color: http://www.k2.t.u-tokyo.ac.jp/vision/dynaflashv2/index-e.htm... And they can now do the tracking without the infrared paint: http://www.k2.t.u-tokyo.ac.jp/vision/MIDAS/index-e.html

The second example doesn't, IMO, demonstrate the kind of deformation tracking the example in the OP does. As far as I can tell, they are able to get a depth representation, segment objects from a video, get an approximation of surface normals and reflectivity for those objects, and project a shaded surface onto those objects.

What they do not seem to be able to do without IR markers is project a diffuse texture to an object so that it would stick properly. See the one example with a non-uniform texture, where the fingers of a hand are fanned out - the texture warps noticeably.

Re: 1000fps image projection on deforming non-rigid surface

#126

Post by that_jojo »

rebuilder wrote: post #125
timdorr wrote: post #8

The second generation of the projection tech does color: http://www.k2.t.u-tokyo.ac.jp/vision/dynaflashv2/index-e.htm... And they can now do the tracking without the infrared paint: http://www.k2.t.u-tokyo.ac.jp/vision/MIDAS/index-e.html

The second example doesn't, IMO, demonstrate the kind of deformation tracking the example in the OP does. As far as I can tell, they are able to get a depth representation, segment objects from a video, get an approximation of surface normals and reflectivity for those objects, and project a shaded surface onto those objects. What they do not seem to be able to do without IR markers is project a diffuse texture to an…

You say that like it's not ridiculously impressive and interesting.

Could you imagine using something like this for reviewing finish options for a product with zero turnaround time?

Re: 1000fps image projection on deforming non-rigid surface

#127

Post by daenz »

goblin89 wrote: post #117
daenz wrote: post #20

Very cool. I do live projection work[0] and latency is always the killer with immersion. Anything higher than a 1000/90ms latency breaks integration at normal dancer movement speeds. 1000fps seems like overkill but it allows for very fast movement. 0. https://youtu.be/ggRcDQZWD_8?t=1281

Is this real-time 3D taking into account dancer’s motions? I imagine it takes a crazy GPU rig? In any case, very impressive.

It is, but I cheat! The dancer has a small android device with a custom gyroscope app, mounted in the middle of her back. I can get her general orientation accurately this way (more accurately, and faster, than state of the art pose estimation).

I am in the process of bringing together a community around art making like this. Let me know if this is something that interests you.

Re: 1000fps image projection on deforming non-rigid surface

#128

Post by balabaster »

Having just watched a bunch of videos on Deep Fakes after the revelation of the Deep Fake video of Nixon's Moon Landing Disaster speech on here this morning, I can't help but feel like this is something else that will make fakes more and more difficult to distinguish from genuine.

Re: 1000fps image projection on deforming non-rigid surface

#129

Post by wajdix »

gao8a wrote: post #109
timdorr wrote: post #8

The second generation of the projection tech does color: http://www.k2.t.u-tokyo.ac.jp/vision/dynaflashv2/index-e.htm... And they can now do the tracking without the infrared paint: http://www.k2.t.u-tokyo.ac.jp/vision/MIDAS/index-e.html

I hope to see this at the next Tool concert

This is what popped in my head while watching the video!

Re: 1000fps image projection on deforming non-rigid surface

#130

Post by goblin89 »

daenz wrote: post #127
goblin89 wrote: post #117

Earlier quoted context omitted.

Is this real-time 3D taking into account dancer’s motions? I imagine it takes a crazy GPU rig? In any case, very impressive.

It is, but I cheat! The dancer has a small android device with a custom gyroscope app, mounted in the middle of her back. I can get her general orientation accurately this way (more accurately, and faster, than state of the art pose estimation). I am in the process of bringing together a community around art making like this. Let me know if this is something that interests you.

It is of interest! I am in the process of learning TouchDesigner, looking to integrate it with Ableton Live. I doubt I’ll have enough GPU power for real-time 3D renders based on changing sound or visual input though…

Currently I am considering pre-rendering scenes to given BPMs (where applicable) and doing only limited realtime alterations with TD nodes.

Post reply on HN