End of the 3D industry as we know?

Mixu

https://www.core77.com/posts/96607/The-NeRF-Method-for-Creating-3D-Models-from-2D-Photographs
What do you think guys? Is it possible that in few years many 3D artists will loose their jobs? Soon or later?

rollin

Jep, exactly this will happen! I'm absolutely convinced and looking for new job opportunities.
This WILL happen!
I mean.. it will happen right after we loose our job because of THIS new invention here: 
https://en.wikipedia.org/wiki/Camera

tynew

This is just high level photogrammetry, which is the direction it should be moving in to aid artists. We still need to clean up the data and for anything that requires movement needs manual work with building/separating meshes. 

This is clearly evident on the ship in the video. I mean it's pretty much unusable except as a reference, because things such as rigging and sails have to be their own meshes if they're to have physics. For basic props though this is awesome.

Did you look at the site? This is the quality of the mesh 



The codes available on their git, now someone just needs to make the software for it. Super useful if you wanna build proper game res meshes around online reference images if it works out?

Obscura

That is gross...

Mixu

tynew said:

This is just high level photogrammetry, which is the direction it should be moving in to aid artists. We still need to clean up the data and for anything that requires movement needs manual work with building/separating meshes. 

This is clearly evident on the ship in the video. I mean it's pretty much unusable except as a reference, because things such as rigging and sails have to be their own meshes if they're to have physics. For basic props though this is awesome.

Did you look at the site? This is the quality of the mesh 



The codes available on their git, now someone just needs to make the software for it. Super useful if you wanna build proper game res meshes around online reference images if it works out?

What if that algorythm will be able to implementing something like quad remesher to that "cubes" mesh? 

Obscura

This is a quad mesh made out of triangles. Marching cubes is just the name of the algo. Its a very common way to poligonize anything that isn't triangles by default.  

The problem here is that you loose a lot of details and it can't recognize separate parts either. So a lot of work is needed to make an actually usable mesh from this. On some environment assets this is fine. Say rocks...  

I do not see how is this useful if the output mesh is this quality. Even standard photogrammetry gives better results.

tynew

Won't make a difference, it's the same as taking that above picture and using quad remesher on it. The entire thing is just soup right now, its the same as photogrammetry. It works for organic stuff like rocks and trees but anything that requires clean and tight angles will need to be pulled apart and scanned piece by piece. I mean just look at those bulldozer tracks, pretty awful.

cptSwing

Ugh, I would not want to work on that mesh. The few instances I've had access to photogrammetry scanned stuff they were absolutely only usable as reference for dimensions etc. Of course YMMV.

Alex_J

I bought this scanned water buffalo (the blue one), thinking I could just drop it in my game. It looks fine from a distance, but when you get close much of the details are just blobbed together. Turn off the albedo map and you can't even tell what things are meant to be.

Took almost ten hours to build a new, clean topology version of it that could be animated, LODed, and so on. But the benefit of that scan is still pretty huge - it saved me time because I didn't need to measure proportions, plus I baked the color map from the scan onto mine as a way to jump start the textures. I still had to do a bit of research and guessing to fill in details from the scan that were too soupy to make sense of.

defragger

Mixu said:

What if that algorythm will be able to implementing something like quad remesher to that "cubes" mesh? 

Ashervisalis

My art director: "Can you please adjust piece A to make it more smooth? Also, piece B is far too angular."

Me: "Sorry boss, this is a photo scan. You get what my camera gets."
My art director: "lol damn"

jStins

Indeed I'm reminded of the time when animators were employed. The time before mocap...

rollin

jStins said:

Indeed I'm reminded of the time when animators were employed. The time before mocap...

Ha indeed!! 
Poor poor animators..

what are all of them doing nowadays? .. Modelers?!. ohh damn.. how can live be so mean to them TWICE??

igi

Depth maps from photos looks convincing. It looks useful for grounding 3d objects in mixed reality. This is not a complete 3d mesh from single photo thing.

NikhilR

Really depends on the end result, this is just one way to get to it.

Also it depends on real world photography so I guess that could change how games depending on realism are built. Still many considerations with optimization.etc, I can see this being more impactful in film than videogames.

One nice thing I'm noticing is that with less actual production process, more time to focus on implementing ideas.

But the scale of a project and the way you invest in it is also a factor.

Like its implausible for a AAA company getting grants from the government to simply belly up if they decide that automation is good enough (which in many cases it is)

Lot of different aspects like image and branding do play into how a company operates and budgets. Like I can tell you that employees of MPC in montreal aren't too pleased about what the company did to the MPC branch in vancouver.

They'll still do their job, but its one of the reasons why there were so many of them at a networking event organized by a rival. 

poopipe

Given that photogrammetry and scanning are widely used already and have been for many, many years this makes a sum total of fuck all difference to anything beyond improving results and/or speed. 

I will be stealing their code though 

sacboi

What do you think guys? Is it possible that in few years many 3D artists will loose their jobs? Soon or later?

not much, no and no...respectively.

Cos, imo, a subd polygonal workflow will likely be standard for a heck of a while longer, still.

Prime8

Actually this will happen at the same time the Euclideon Inifinte Detail Engine replaces all polygon based engines ...