Hacker News Reader

Comment by mike_d | Hacker News Reader

From the demo video this looks really interesting for games/use cases that are centered around the >1000ft view.

Even in the video showing off the technology things look downright terrible up close. If you pause at the very beginning you can see handrails on the stairs that are just odd floating blobs, cars have all melted into the ground.

It is incredibly impressive that they have managed to get aerial LIDAR into a game engine, but the visuals are quite jarring up close.

View on HN

This is due to the method of data acquisition. My company also uses cesium (in the browser) for visualization at sub cm level for inspection. Our quality bar is that inspectors have to be able to inspect the grain of the grout between the bricks.

We acquire the data in 3 modes for high buildings, we do low altitude fly over with the 2m drone, then we do facade with the regular drone, and finally we shoot from the ground. We only do ground based lidar, but aerial lidar is improving so maybe that's next. The rest is all high detail photos fed into photogrammetry.

The final results look perfect in cesium, some thin rails might not show up, and windows get a bit funky, but the geometry and the textures load in without a hitch, really cool technology that's so simple in the basis that we built an alternative viewer in Rust+WASM in 3 weeks.

Wait, if the technology is so simple that you can build an alternative viewer in 3 weeks, what are they selling? Is it the data format?

They’re selling it so you don’t have to get it by yourself.

3 weeks for a prototype

3 months for an alpha

3 quarters for a production release.

You know how engineering estimates work... ;-)

They've got a whole bunch of features in their viewer, making a feature matching viewer would take years. The basic technology is a tree of JSON files describing what models or other JSON files to load for regions of space. So the nodes are JSON files and the leaves are 3d models basically. There's a simple algorithm that you give your current perspective (i.e. where you are in space, and what direction you're looking in) that will walk the tree and decide what models will be in your field of view and which won't be, and at what distance (which decides the level of detail). Then the last step is simply downloading those models, while at the same time evicting irrelevant 3d models from your gpu.

They hand rolled an entire high performance 3D engine in Javascript, which is a big challenge in and of itself. We cheated by using WASM and a high performance programming language, making the 3D engine part trivial. Likewise, I'd be surprised if it took them more than a week or two to get their initial proof of concept working in Unreal Engine.

Oh and I believe their business model is/was to sell hosting and processing for the 3D models. We've been paying them a couple hundred per month just for this service.

That's helpful, thank you!

Off topic, but: would you mind telling which software you are using for the photogrammetry?

Having worked at a company with an identical premise, Reality Capture (because it's faster) or Bentleys

Nice! where did you work?

We're using Metashape and Reality Capture. We used to use Bentley, but they've got unethical business practices so we're avoiding them now.

Metashape has really good licensing, and their software runs on Linux and is easy to integrate into processing pipelines, also they've got a big featureset.

Reality Capture was super expensive, though that has seemingly changed now that they're bought by Epic, so we might reconsider our strategy there. They make better use of hardware, making them faster, and their resulting models look great. Their licensing was so expensive that we just couldn't afford switching all of our processing to RC.

> We used to use Bentley, but they've got unethical business practices

Go on. I'm curious now.

You can google a bit about Bentley Connect charges, we were not the only ones who called their sales reps in a panic.

https://community.spiceworks.com/topic/2129583-gathering-inf...

If you can get a good deal with them it's probably fine, and we only talked to nice people there. But we're still on limited runway, we can't deal with random $20k invoices, even if a nice sales rep reverts the charge when called.

Could you use a technique like this to do facade inspection as mandated in nyc?

I just happened to watch "how to with John Wilson" S01E02 which was talking about how scaffolding is put up everywhere in nyc because the facade needs to be inspected for loose bricks every 5 years

I imagine this will be used for stuff outside of games primarily. My understanding is that Unreal is now used a lot in specialized domains like film-making (globe fly-bys in the news and documentaries?) and sort of general "dashboard"-y stuff.

I'm sure someone could think of an "interesting" game built off of this data but I highly doubt you go through all this engineering effort on that idea alone.

> Unreal is now used a lot in specialized domains like film-making (globe fly-bys in the news and documentaries?)

It goes further than that now. For The Mandalorian, they famously set up a "20-foot high, 270-degree semicircular LED video wall"[0] to use as a backdrop for the shots, and that wall was rendering the desired environment using Unreal Engine. No need to chroma key an alien planet in post, when you can make it a part of the scene (and now the actors can see it too as they play!). More at [1].

[0] - https://www.polygon.com/tv/2020/2/20/21146152/the-mandaloria...

[1] - https://www.unrealengine.com/en-US/blog/forging-new-paths-fo...

ILMVFX's YouTube channel has some fantastic BTS footage of how this works, for anyone curious:

https://www.youtube.com/channel/UCjnYk44Aj9E634TPucpIXnQ

From what I read about it previously I think they do actually re-render the backgrounds in post

But the LED video walls serve two purposes: ensure the lighting and reflections on real objects in the scene match up with the CGI, and give the actors a visualisation to work with.

> From what I read about it previously I think they do actually re-render the backgrounds in post

All the videos and reels I've seen on this talk about "capturing the effects in-camera", so while I wouldn't be surprised if there's some integration and cleanup work to be done (particularly where the wall meets the set), it does sound like what is on the LED wall is what we see in the final footage and it's not just placeholder or pre-viz.

I think you're right. They use terms like "final pixel" when they talk about it, which leads me to believe they're capturing the background and not just rerendering in post.

I think the biggest wins they get from it are that the actors can see what they're supposed to be standing in front of and that they get realistic lighting on the actors and other props from the environment

It can also be used as a starting point to "rough in" your layout and then clean up and refine by hand..

If you were trying to build a virtual version of a real-world location, I would imagine the time savings in starting from real landmarks and buildings would be significant.

> My understanding is that Unreal is now used a lot in specialized domains like film-making (globe fly-bys in the news and documentaries?)

It's also used to create TV studios [0]. In the UK, BBC soccer punditry programmes such as 'Match of the Day' are filmed in a green-screen studio with the pundits, desk and chairs, and a virtual studio created using Unreal Engine. Different programmes can get a totally different look by using different Unreal environments.

[0] https://www.unrealengine.com/en-US/industry/broadcast-live-e...

Also good for prototyping and seeding ideas for level design, I'd imagine. I saw someone proto their run-and-gun game on google maps geo - they'll have to model everything once they lock a location down, but can test complex environments very quickly. Also having a real location, however rough, as a base gives you a lot of ideas you would have a hard time coming up with, starting with a blank slate.

It would still lend itself to some interesting games, such as Mech Battles or Warcraft with oversized vessels or hordes of leming-sized people in these environments.

Maybe a game of Global Thermonuclear War?

Shall we play a game?

It has very distinct low-res photogrammetry look yes. But if embraced and under good art direction I think it could be made to work as a stylized look. Maybe throw in some shader effects to spice it up a bit and make sure that the hand-modeled objects fit into the design, and it could be pretty cool.

I've noticed a trend of games graphics being fairly oldschool while the mechanics are modern. Maybe this'll lend it's self to that trend?

Reply on news.ycombinator.com