• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Siggraph Asia: Create 3D editable models with a single photo

Status
Not open for further replies.
As a mathematician I'm not too impressed, but certainly nice. Given what they're showing I'm actually surprised this hasn't been done before. Doesn't really seem too groundbreaking (regarding the actual mathematical algorithms/procedures). What's certainly impressive though is the supposed accuracy and simplicity of the tool.

I'm also a bit unsure how they managed to get parts of the background for the telescope picture that were not in the original image. Patchmatch algorithm? Sure. Seems pretty impressive (at first glance) though. The failures though are what I'd expect. Still, really nice work.
In hindsight you're probably right because in theory all the ideas suggested exists and works, it's just that we haven't seen it done in practice. A siggraph presentation also allows them to show the best cases as an initial starting point for a paper.

All those dick jokes aside, I really hope to see this eventually filter down to make this stuff more accessible to more people.
 
They show how they get the background. It's the same algorithm as Photoshop's content-aware fill.

Overall this is a really cool simple tool doing a bunch of really obvious stuff that I'm surprised no one has done yet.
 
Fuck...this is revolutionary. Imagine how would it speed up assets creation in video game industry.
 
Create a new type of captcha where instead of correcting OCR text users correct outlines automatically read by the software to create accurate, symmetric models. You have now farmed out all of your basic object modeling.
 
A whole new generation of "... Ate my balls" comics will flood the Internet.
That's funny, I was actually just reading Mr. T Ate My Balls last night for the first time in years. It'll be the new Pictochat! This is one o those things that I see and think to myself, "Why didn't I think of that?".
 
This is pretty clever, especially the elegant user interface.

However, this method has its limits in that the more symmetrical a shape is, the better it works. And I expect it to break down completely for random shapes like rocks for example.

Still, awesome demo.
 
I read the title as Create 3D edible models with a single photo.

I am slightly disappointed.

Lol, yep same. I was thinking it would be similar to that thing that "prints" a scan of your face on top of a coffee or something but you can still drink it.

This is really cool though. Limited... yes, but should be easy for people who can't model in the slightest.
 
I could see this being very worthwhile for prop making in video games as long as the models that are created arent horribly unoptimized.

Create a new type of captcha where instead of correcting OCR text users correct outlines automatically read by the software to create accurate, symmetric models. You have now farmed out all of your basic object modeling.


Clever girl
 
Game devs should use this to easily make models of symmetric objects. Will save a lot of time and manpower I would imagine.
 
Well, this is pretty awesome.

It's no magic bullet as far as 3D modelling is concerned.

Certainly, not if you want game quality assets out of it.

But damn if it's not a quick way of mocking up a 3D that you can adjust and enhance.

And fan-fucking-tastic for advanced image editing in photoshop.

Which is the real prize here - that 3D tech can be incorporated into 2D image manipulation at a very low-cost (training and usage) for most users now.

Also, a fantastic way of creating a massive world wide object recognition database.

With other advances in AI tech and with a little bit of human cognitive processing power to aid, the technology to create a full-fledged, 3D mirror world isn't too far away at all.

Moreover, we can have that mirror world updated on a regular basis, so that we can actually search through the mirror world along a time line - essentially time traveling.

And this tech will allow us to extract 3D information easily from any photographic data from our past. The ideal here would be a VR-enabled Google maps that you can explore at any point along the time line for which digitized data exists of it, in high fidelity - afforded by AI driven 3D extrapolations of image/object data from photos.
 
Neat stuff, but as just stated before me, it's not at a point where it's useful for anything outside of quick preview use.

This is pretty similar to how SketchUp Building Maker works, and I doubt it was first seen there.
 
Status
Not open for further replies.
Top Bottom