3D Gaussian Splatting - Why Graphics Will Never Be The Same

Thank you for not turning this into a 58 min long documentary.

awesome and hilarious. my favorite combo.

Now, on to use case-specific optimizations, the main issue might be capturing, storing, accessing, and displaying this kind of data to moving objects in the scene, so that we don't have an AR style effect occur that would destabilize the immersion

absolutely unwatchable because of 'inhuman' narration, no pause, no emotion, nothing.. is this the future ?
like chewing a paper in the noble restaurant, absolutely disgusting these 'nerd' times

Thx for saving my time.

What’s the advantage of using Gaussian splat? Great video but I’m slow at learning this 😂

Isnt Nanite better?

i guess we're not ready for this.. but our kids are gonna love it

I wonder if this will have the same problems that vocal rendering has. What will it be like to animate? Or to deform?

google maps is about to be insane

i imagine this'll have its applications in things like street view or what have you but i don't think it'll find its way much further than that

bill wurtz and acerola combined into one

knock knock it's religion

oh shit wrong channel

Kinda feel like I'm looking at hyperrealistic sprites, feels weird

so it's like if voxels were the size of a pixel

At some point graphics are going to become so real people will get PTSD from playing violent video games

what ever that neural network or ai is doing it looks uncanny af.

why game optimization will never be the same. oh, we released a remake of a 1999 game that has one location and a top-down view, if you don't have a 4090 you can go to hell

These ideas aren't worth anything.
The application is everything.
Just like the iphone in 2007. It wasn't the first.

this has to be essentially built from the ground up meaning we're like what 5 years maybe from it being useable probably?

I don’t get it

This works ok for still images, but when you have a scene that has lighting changing, lots of animation, etc... not as much.

welcome back bill wurtz

okay, but show me a game with it

vram requirements have been traveling the exponential curve

For anyone confused by this and it's feeling short. In a nut shell you take a few flat 2d photographs of an area then it gets processed into a fully 3d scene with some machine learning filling in the gaps based on the pictures.

Welp nothing yet

I love that it all just feels like sarcasm the entire time

So... it's just 3D point cloud. This could never do real time lighting unless i'm missing something?

Real life graphics explained by a real life robot. Thanks. I'll go cry in a corner muttering words of glorious days past.

Are you eminems brother?
These ad libs man xD

This is perfect for when you want to buy a house, retail companies can let you download a walk around gaussian splat sim for you to explore.

wow it looks beautiful

Heavily reminds me of “the history of the entire world, I guess”

If it can be dynamic and not need a ton of vram for players then maybe it's the future

A company in Australia called Euclidean figured out how to rasterize point cloud 3D data at high refresh rates years ago, but nobody wants to develop a game using point cloud data let alone this which is even more complex.

Hey, you may have missed it, but your analogy about Gaussian splats to neural networks is a revolutionary way to represent neural networks. Neural networks are simply a high dimensional image, and using N-Dimensional Gaussian splats would dramatically speed up neural network implementations and make them far more precise and detailed. You just exposed a new form of neural network technology that will change the world.

sweet but how much data is involved?

Wow, so much technology, and still can’t render first-person legs

yeah, by now iam shure we live in a simulation

Masterclass in explanation and editing.

Doom added shadows?

Crime scenes, mapping, etc.. But not video games or movies, no.

Me, watching the video and still not even getting git.

Ah, so that's how unrecord did it

You lost me at splat.

I'm here one year later. And i just wanna talk about how disappointed i am that this changed nothing. Everything is still the same. What went wrong?.

Well, it looks more like a niche application.
It can't store logic. It doesn't work with objects. It can "create" an image, it works on existing ones.
How do I see it?
1. You render cool models and scenes from a shitload of angles
2. You put it all that into gaussian splatting
3. You can move and explore an extremely high-quality model realtime

Would be great to see street viewing like that on Google Maps.
Perfect opportunity for websites to look even cooler on all devices (using pre-made scenes).

Maybe there'll be a new era of 3D-videos.

New tech like this won't be standard for like 20 years