The cathedral is the star of the show, but I found myself more fascinated by all the incidental little details captured in this. Like the trees and the buildings nearby, the basketball court, the little solar panel on the side, the apartment towers.
It's like seeing the promise of Google Street View realized.
The technology is in such an early stage and it already slaps this hard.
Wonder what the next steps are for this tech? "Un-splatting" optimizations to convert things like flat surfaces into classic rigid textured geometry? Factoring the lighting away from the geometry? Filling the pipeline with AI tooling - segmentation to remove the "variables" like humans automatically, generative AI to fill in the missing non-semantic details based on the surroundings, more generative AI to "port" the data from things like public high resolution images of plaques or other undercaptured details into the scene?
Seeing this kind of reconstruction now makes me very excited for what the 3D environments of the future are going to be like.
One other little easter egg you may not have noticed. There's an American flag gently fluttering in the breeze on top of a nearby building. We were discussing whether to adds some birds as well, but that never made the final cut. You ask about factoring lighting away from the scene - actually, PlayCanvas can already do relighting and shadow-casting for gaussian splat-based scenes:
Some kind of day/night toggle would have been another great feature...but again, it didn't make the final cut of the experience. Maybe a HD Remix is needed. :o)
> This scan is made possible by recent advances in Gaussian Splatting. This is an emerging technology that lets us quickly create very detailed models just from photographs. For this model (or splat, as we call them), my friend Daylen and I flew our drones around Sutro Tower at a respectful distance for an afternoon until we had collected a few thousand photographs.
> I then aligned these pictures in free software called RealityCapture. Alignment is the process that teaches the computer that a bunch of points in different images all actually correspond with the same point in real life. Then I used another piece of free software called gsplat to produce the 3D model itself.
Yes, I also added a bit of context on twitter (https://xcancel.com/fulligin/status/2079296985811873796) and will probably blog about the process a bit more. Grace was a lot more work than Sutro. I wanted to be ambitious in combining interior/exterior scenes in a new way, and I also wanted it to be less of a "toy". Grace actually plans to use this to show prospective renters the space, and we spent a lot of time making sure it could run well on a wide variety of devices.
In case it's useful, given your last sentence, it runs really well on Chrome on Android (on a Pixel 10) but it's a bit glitchy (although not impossible) on Firefox on Android. The loading indicator on Chrome got stuck briefly whereas the Firefox one was smoother, but that's about the only case there Firefox is better.
I think it's probably more like an "incorrect" focal length for what you'd expect. And it's kind of hard to know what you'd expect because you don't typically have arbitrary spatial control over your point-of-view, so some POVs you're used to only seeing through cameras (or even certain types of cameras), while others you're mostly used to seeing through your own eyes.
Reminded me of a VRML demonstrator inside Notre Dame from decades ago. Different approach, similar outcome. (it used surface texture mapping inside a game engine from my memory)
A lot more architecture would benefit from being available like this splat model. The Pharonic tombs which you can visit in google/other virtual tourism are good, but this is also good!
Did you animate the cars separately? That really did a lot to add to the immersion (and the sound effects too). If nothing moved I think the effect would be significantly diminished.
Yes, the cars are separated out and procedurally animated. They're splats, like the rest of the scene. Easily done in PlayCanvas - simple but effective!
I spent a lot of time hand-masking out inconsistencies between shoots. I waited too long between sessions and they actually built out an entirely new section of ceiling. PPISP guidance in splatting also helps a lot in solving sensor inconsistencies between the many different cameras and lighting conditions that went into the source dataset.
Extraordinary. Thank you for sharing. If there would ever be the possibility of seeing European Gothic cathedrals in this manner, that would also be really cool.
Hi Vincent, the Sutro Tower 3dgs demo is one I enjoyed a lot. I'd whip it out on my phone and show people saying, 'hey look this is whats possible with 3D Gaussian Splatting'. And this Grace Catherdral splat is stunning. It showcases one of my favorite parts of 3dgs which is reflections and glass. The color and essence of the stained glass comes through so well.
The cutaway Peek feature is like a sci fi version of an Stephen Biesty's Incredible Cross-Sections book. Incredible work!
The Keith Haring altarpiece is a really cool surprise!
It's unfortunate that this tour makes it very difficult to see. Really not a fan of this "on rails" presentation style. I understand the angles matter when viewing a splat, but come on.
It's like seeing the promise of Google Street View realized.
The technology is in such an early stage and it already slaps this hard.
Wonder what the next steps are for this tech? "Un-splatting" optimizations to convert things like flat surfaces into classic rigid textured geometry? Factoring the lighting away from the geometry? Filling the pipeline with AI tooling - segmentation to remove the "variables" like humans automatically, generative AI to fill in the missing non-semantic details based on the surroundings, more generative AI to "port" the data from things like public high resolution images of plaques or other undercaptured details into the scene?
Seeing this kind of reconstruction now makes me very excited for what the 3D environments of the future are going to be like.
reply