Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

vincentwoo.com

Post reply on HN
Print view View original

181–190 of 198 posts

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#181

Post by skyyler »

xipho wrote: post #102
skyyler wrote: post #57

Earlier quoted context omitted.

No. It is the good weather capital of the country. Mild winters, mild summers. Not too much rain. No serious threat from tornadoes or hurricanes. That’s a very big draw and it wouldn’t go away by making more dense housing, even if the rest of the peninsula was developed like Manhattan.

And it lives on a series of incredibly active fault lines. During my undergrad I had multiple geology profs adamantly mention that when not if aspect of this, and on time-scales of 100s of years, not 1000s. YRMV.

A lot of people would take earthquakes over hurricanes, tornadoes, blizzards, hail, 120f summers, -20f winters, and more.

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#182

Post by ladon86 »

poutrathor wrote: post #169
cloudfudge wrote: post #167

This brings a meta quest 3s to its knees. It's almost so bad you can't quit it, and the video lags 15 seconds, which is very disorienting to be immersed in (it can make you fall down). Shame, since it looks gorgeous.

what part is the culprit : the meta quest or the 3D implementation from the website ? On a classis laptop, it behaves well

The Quest's Snapdragon GPU, like most mobile GPUs, uses a tiled rendering [1] architecture.

The basic technique for rendering gaussian splats is kryptonite for this architecture, essentially implementing every worse practice for rendering on a mobile GPU:

* Tons of overdraw (overlapping splats)

* Tons of alpha blending

* Millions of splats in the distance generate a lot of tiny triangles resolving to a single pixel

* Long thin splats in the foreground generate triangles that cover multiple tiles

These are all the ingredients you need to bring a mobile GPU to its knees! Any desktop GPU (including most laptops) will be far less sensitive to these issues, even if it's not very powerful. It's a fundamental issue of architecture rather than one of raw FLOPs.

[1] https://en.wikipedia.org/wiki/Tiled_rendering

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#185

Post by diogocp »

akanet wrote: post #143
atarian wrote: post #19

very cool, reminds me of watchdogs. where's the music from? it's very calming

It's actually a mix my girlfriend made, but the main track is https://soundcloud.com/martinlandh/moonlit-serenity

What about the track in your video "I finally made it to the top of Sutro Tower"? It's beautiful!

The video: https://www.youtube.com/watch?v=A7-3CKulsCc

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#186

Post by akanet »

zokier wrote: post #161

What is the limiting factor for level of detail here? Is it the source data (too low res camera/too far away), the processing, or getting it rendered in real-time in browser? Is this the full detail, or is this somehow downsampled/compressed version for the web?

The processing and alignment of source imagery, imo

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#188

Post by jg0r3 »

akanet wrote: post #98
armchairhacker wrote: post #92

How feasible would it be to generate high-quality guassian splats of everywhere, starting with big cities, like Google Maps street view and 3D view? Then you can make video games and interactive experiences in real-world locations. I doubt the collision handling is there, but you can at least start with something like Microsoft Flight Simulator with low-flying drones. It would also be great for training AI to generat…

the collision handling part is actually not that hard to back into from the splat data. i think HQ splats of everywhere is something we'll start seeing in the next decade! we haven't quite solved scaling issues, LoD, and streaming for splats but the early work is promising

Could you direct me to resources about collisions and splat data?

I have an extreme interest in deriving measurements from splat data of vegetation, as it tends to reconstruct thin planes like leaves far better than other traditional SfM techniques.

Re: Show HN: Immersive Gaussian Splat experience of Sutro Tower, San Francisco

#190

Post by scoofy »

dragonwriter wrote: post #148
scoofy wrote: post #139

Earlier quoted context omitted.

This is a general argument assuming units being arbitrary. Units should be effectively arbitrary, but every town will have different rules.

> This is a general argument assuming units being arbitrary. Well, no, it doesn't assume units are arbitrary, it assumes units are fixed square footage, which they are not. Under most regulatory schemes, there is a practical minimum size or a habitable unit, but a pre-existing area zoned for detached single-family units exclusively is unlikely to be comprised of single-family units that happen to also be the minimum…

The same concept, at a minimum, would need to be extended to units. This is what I mean by assuming it’s arbitrary. It’s just redundant to say that, yes, obviously you need an incremental increase in units, sqft’s, footprint, vertical footprint, etc.
Post reply on HN