A Revolution in Computer Graphics Is Bringing 3D Reality Capture to the Masses | Hacker News Reader

A Revolution in Computer Graphics Is Bringing 3D Reality Capture to the Masses

singularityhub.com

45 pointsBrajeshwar17 comments

Photogrammetry capability has been around for decades. It's faster now, and the auto-alignment works better.

Open Drone Map does much the same thing, by different means. You feed a few dozen pictures from your drone into the program and get out something that looks like Google Earth. Open source, too. Machine learning techniques have helped the alignment and camera calibration problems substantially.

It's still a niche thing, like spherical photography.

What's a niche thing today can become a common thing in the future, though.

For example, exposure bracketing and stacking to get HDR images had to be done using dedicated software fifteen years ago or so, but nowadays most cameras and phones do it without any additional steps.

Good example. Image stacking too.

Why are the free phone scanning apps so far ahead of their open source competition? Is it just the iPhone's lidar, or do these apps work just as well on Android?

Because I've tried Meshroom, and while it has a surprisingly nice UI, it took my desktop 3 hours to compute a relatively mediocre model of a tiny part of my hallway.

I would really love an improved LiDAR Scanner on the iPhone that can easily incorporate it into taking pictures. It would also be amazing training data for LLMs to generate 3D models from a description.

I don't know why Google's Pixel phones don't have depth scanning like iPhone's but instead added ... a thermometer? Is there a patent issue or something?

Is there a good iOS option for 3D capture of objects for manipulation in AR? Or at least just the object capture (eg rotating an object to capture it's faces)

Apple has released a pretty polished object capture app for ios. https://developer.apple.com/videos/play/wwdc2023/10191/

This looks cool, but your link is to a video describing it, and the app “Object Capture” doesn’t appear to be in the Apple App Store.

https://developer.apple.com/augmented-reality/object-capture...

Though it's mostly sample code.

Which apps are good to try on iOS?

Both apps mentioned in the article are available for iOS.

The tweet quoted in the article referenced using “ iPhone 12 Pro + @Scenario3d” but that account doesn’t exist and I don’t see any results online for it. I’m curious, too, because the scan looks maybe better than polycam to me.

https://www.reddit.com/r/photogrammetry/comments/x12gcp/intr...

PolyCam is apparently not free but considered one of the best. I used 3D Scanner from Laan labs (& it's a free app. No ads thankfully). I used it to build a few dataset for NeRFs based on MIPNeRF. The data quality seemed pretty fine. I use a iPad Pro with M1 btw.

https://apps.apple.com/us/app/3d-scanner-app/id1419913995

Confirmed. I have tried all photogrammetry apps on the iPad and political is by far the best. Regardless of how good these apps are the meshes still need a hell of a lot of treatment before they can be used in production.

I'd be curious to see a comparison with Kiri Engine and see how that compares to the alternatives, and with the various different methods supported by Kiri Engine.

Reply on news.ycombinator.com