KIRI Engine - 3D Scanner App

Honestly, PBR generation from 3D scans has always felt like a compromise. We trained a diffusion-based model to better separate material properties from scan data, helping reduce noisy artifacts and generate cleaner PBR maps. The difference is pretty obvious. Check the comparisons below.

@torisetxd The lamp was physically there during the scan. Here's the picture of lamp. It moves around but the lamp and chair are always together.

@KIRI_Engine_App this looks really like hallucinations, that lamp was a pretty incredible guess

AI-Enhanced LiDAR. Left vs right. A real LiDAR device costs thousands of dollars. What's in your iPhone Pro is a "baby LiDAR". Limited depth resolution, noisy output, not really built for high-precision 3D. You can't change the hardware. So we built an ML layer on top. Denoising, geometry completion, detail recovery. Processed server-side. Same sensor. Very different result.

@AremasNL LiDAR scanning is local, but our upscaling is a pretty heft process so it's cloud based

@KIRI_Engine_App Is this Cloud or local?

@torisetxd Yeah it's just using the actual photo info so it doesnt really have to guess much other than depth.

@AkshayPL Not yet but were looking at using the methods with other scanning types.

@KIRI_Engine_App Can it use RGB data to estimate the geometry?

@_____Lightning Were working on expanding it to other areas 😄

@KIRI_Engine_App Did you combine this with anything else, such as photos or this is just pure lidar?

@530RB The lamp was always there. Raw LiDAR just couldn't see it.

@KIRI_Engine_App But where did the lamp come from

@hone1er Possibly! That's honestly a question for scanner makers - were trying to keep things cost effective and phone based.

@KIRI_Engine_App Makes me wonder what this tech could do combined with an actual purpose built 3D scanning device 🤔 are they already good enough without it? Can the cost be cut in half? Speed of scans increased? Love the phone aspect though. I use it to scan moped parts for 3D modeling

@Mrfunnysheep Were looking into expanding the tech to our other modes 😁

@KIRI_Engine_App I wonder, if this “Upscaling“ would also work with low confidence depth estimation from motion, as an alternative for non Lidar equipped devices.

@AkshayPL That's exactly the direction we're heading. AI-estimated depth as the input, then enhance from there. Pipeline is in progress. Stay tuned.

@Mrfunnysheep Similar idea, instead of raw depth from motion, we're looking at using an AI model to estimate depth first, then run it through the same enhancement pipeline. That's actually something we're actively building. Stay tuned, should be a few months out.

@zeelee3D Oh that would be awesome! I think I remember someone filling potholes with mosaics but it was a loooong time ago, you should definitely do it!

@KIRI_Engine_App You know... way long ago I had the idea to do this with mosaic ceramics in my city. I was way too lazy to actually do it, but this is pretty cool. Especially if we can get some natural print materials.

Guerrilla urban repair, powered by 3D scanning. 🛠️ We found a creator using KIRI Engine + 3D printing to reverse-engineer and perfectly patch up broken steps, walls, and ledges around their city. We're huge fans of this kind of work and want to help this small channel grow. Check the video out👇

@Almorgand Woaaaaah just checked out that dudes stuff - that's like god tier whatever the name is for this type of graffiti

@KIRI_Engine_App You need to contact instagram.com/3dfiti !

@taziku_co It's so great. Such a combination of art and technology in a way that's easily understandable and enjoyable by everyone 😄

@KIRI_Engine_App これ、3Dスキャンの実用先としてかなり面白いですね。 都市の隙間が新しいアートへ！日本でも是非やってみたい。

Check out the video here: youtu.be/n5VUG7CZ860?si…

Apple just trained a 3D Gaussian head reconstruction model on 10,000+ subjects. Feed-forward. No test-time optimization. New identity in, reconstructed Gaussian head out. The UV-parameterized Gaussian representation decouples the number of Gaussians from the number and resolution of input images, making it practical to train with many high resolution views. And the heads are not just static either: text-conditioned identity generation, plus blendshape-driven latent animation across identities. apple.github.io/ml-headsup/ We've been building in the 3D Gaussian Splatting space for a while. The gap between "research demo" and "works on real people at scale" is closing fast.

@DanyBittel Yeah man, here's a link 👍 It came out very much on the more transparent side though. So awesome to play with your data man. Really big, big thank you - literally everything you scan I can feel the immense amount of time that went into it kiriengine.app/share/3dgs?tas…

@KIRI_Engine_App Can I look at the 3dgs somewhere?

The undisputed king of micro splatting @DanyBittel has generously shared a bunch of his data. Go check it out now, it's really an amazing gift! We threw it into KIRI Engine to check how our 3DGS to Mesh 3.0 would perform on a micro level and as hoped, good data in = good data out 😆

@Almorgand Thanks!

@KIRI_Engine_App Great stuff! Super clean.

Honestly, PBR generation from 3D scans has always felt like a compromise. We trained a diffusion-based model to better separate material properties from scan data, helping reduce noisy artifacts and generate cleaner PBR maps. The difference is pretty obvious. Check the comparisons below.