Skyfall-GS – Synthesizing Immersive 3D Urban Scenes from Satellite Imagery

"explorable" and "immersive" is definitely a bold choice of words when you can't really get below the level of the buildings before the gaussian splatting is very obvious. Sure, it's impressive that you can get that detailed from a few satellite images, but I think that might be overselling it a bit

While the methodology wasn't published, Microsoft did something similar for Flight Simulator.

https://www.smithsonianmag.com/air-space-magazine/flight-box...

It looks good! I imagine a reasonable next step might be to do something about the cars, which are omnipresent in urban scenes but seem like they've been left a blurry mess in the examples.

Maybe dumb question but how do I just take a sat image and create the scene? The scripts in the repo are all about training which I assume requires you to have the 3d data too.

These sort of projects always look cool but I think the real "wow factor" would be a file upload where you can see the result on your image. I assume there are reasons why this isn't done.

This is so cool. I used to work on urban heat island analysis and now work in natural catastrophe modelling, and in both cases knowing the average heights/volumes of buildings is a very handy thing to have but is surprisingly difficult information to retrieve. Even a coarse estimate available at annual resolution has some really awesome use cases, very excited to see this.

Is there any reason that this couldn't integrate Street View data?

This would be the next step for flight simulators, which while remarkable still require handmade assets for accurate details.

Very cool; interesting how it turns all the trees into puffballs though. Some artifact of the pre-trained depth estimation or diffusion model maybe?

This could be specially good for a world 3d model for flightgear.

Now the GTA: Anywhere please...

Nice, but when you look up close things like this and Google Earth look like a post-apocalyptic scene :)

It would be amazing if they could also take user-generated photos and videos at ground level and accurate mapping data (that has building outlines) and clean that up to something presentable.

I mean, what they do here is what google and apple are already doing for years. It's time for the next step.

Re: utility in games,

I suspect hybrid solutions will remove the limitations of GS, with (eventually...) some smooth hand off. Do clean-enough GS like this; then hand the output to other systems which covert into forms more useful for your application and which adopt e.g. textures from localized photos etc.

It's just a bit of engineering and compute...

[flagged]