Voyager: Real-Time Splatting City-Scale 3D Gaussians on Your Phone
arxiv.org

・

54 points

・

PaulHoule

・

16 days ago

18 comments

reactordev ・ 15 days ago

Been doing this at Faro Inc since 2023 - I helped build it. The real magic is simply the lookup rasterization on device. Since mobile device GPU’s are fast now it fits inside the geometry shader.

  • tobinc ・ 15 days ago

    Any word if Faro is working on anything like Leica's Powerlock for laser trackers?

    • reactordev ・ 15 days ago

      Faro does scanning, not tracking. It shoots lasers in all directions while simultaneously taking 360deg imagery, resulting in high density colored point clouds and gaussian splat pre-imagery. I no longer work there as they up rooted their executive team.

jauntywundrkind ・ 15 days ago

Are there any existing examples of partial render offload to the cloud?

Crazy good insight here: splatting is largely a search problem, and that can be offloaded to the cloud.

> Specifically, on the cloud side, we propose asynchronous level-of-detail search to identify the necessary Gaussians for the client. On the client side, we accelerate rendering via a lookup table-based rasterization.

ge96 ・ 15 days ago

Why am I seeing "splatting" more often recently thought it was old tech

  • littlestymaar ・ 15 days ago

    I don't know what was the use of “splatting” before that, but the modern trend has started just two years ago with this paper: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    • corysama ・ 15 days ago

      Various forms of point/blob rendering have been around for decades. What has been missing has been good workflows to create the content.

      That paper kicked off a rapid stream of a thousand papers by taking a photogrammetry-style workflow and producing better than photogrammetry results by reframing the process as gradient decent on differentiable point samples. This allowed the research to stand on the shoulders of all the work being put into deep learning tech.

      • IshKebab ・ 15 days ago
        ・ 2 more

        This is like saying the car is nothing new because various forms of wheeled transport have been around for centuries.

        Typical HN.

  • hirako2000 ・ 15 days ago

    If you call a few years, 'old'.

    Anyhow the novel approach opened the route for building on top. A paper recently proposed dynamic captures, for real time animated Gaussian Splats.

    It will go on there is so much to explore research-wise, optimisation like this innovation along is a large field of efforts.

  • boulos ・ 15 days ago

    You're not wrong.

    The early 2000s splatting from point-based rendering is what George Drettakis and his students realized could be applied in this new NeRF domain.

    Basically, all the reasons that point splats didn't work for regular surface rendering nearly 25 years ago (holes, inefficient, can't edit them like meshes) are less of an issue in a light field capture setup.

  • nomel ・ 15 days ago

    Most tech is "old", with new use cases and accessibility. I think it's most interesting when it jumps into my pocket.