3DGS, 4D Gaussian Splatting and Beyond


Гео и язык канала: Весь мир, Русский
Категория: Технологии



Гео и язык канала
Весь мир, Русский
Категория
Технологии
Статистика
Фильтр публикаций


Репост из: Physical AI
Видео недоступно для предпросмотра
Смотреть в Telegram
Splat too heavy to open? Drop the giant PLY into your browser and get a streamed SOG with real LODs, ready for SuperSplat. No servers. No installs. Open source.

https://sog.flyreelstudio.eu/en/ Your files stay with you: nothing is uploaded, everything runs on your own GPU.

Open source (MIT): https://github.com/AndriiShramko/shramko-sog-lod-converter
better quality on my YT https://youtu.be/VNLFXHHV094

Then go flying: the best 3DGS FPV sim → https://gsfpv.flyreelstudio.eu/en/

#GaussianSplatting #3DGS #SuperSplat #WebGPU #OpenSource LinkedIn:

Lets discuss your project!
https://lnkd.in/gq_RenVs
Book a call: https://calendar.app.google/Ff729HqGk4RpzPNDA
Email: zmei116@gmail.com

Source: https://www.linkedin.com/posts/andrii-shramko_gaussiansplatting-3dgs-supersplat-ugcPost-7513565161815719936-9sNa


https://louenpottier.github.io/lagsplat.html
LaGSplat infers the physics of an object from a few seconds of video: a visual encoder projects each frame into a low-dimensional latent space, a Lagrangian Neural Network learns the equations of motion there, then a Gaussian Splatting decoder reconstructs the scene in real time. An external force can be applied at any instant, anywhere in the scene, even though the training data contains no measured force.


Репост из: Physical AI
Видео недоступно для предпросмотра
Смотреть в Telegram
Source: https://www.linkedin.com/posts/spenserdickerson_gaussiansplatting-3dgs-vfx-ugcPost-7512985140634177536-RcdX


Today I launched SplatPaint 1.0, a procedural editor for Gaussian splats that runs entirely in your browser.

Gaussian splatting is changing how we capture the real world in 3D, but the assets have been hard to work with once they're captured. SplatPaint helps to solve that. You can paint, stencil, relight, sculpt and deform splats, then push them through 38 stackable real-time FX, all without installing anything.

The 3-minute walkthrough below goes from the site straight into the editor: relighting , painting and grading it, projecting stencils, sculpting with a lattice, then running some FX, fire, particles and hologram FX all in real-time completely built from splats.

To celebrate the launch, SplatPaint is free for everyone for the next 7 days, until Monday, October 12. No signup needed!

Check it out:
https://splatpaint.app/






vk_gaussian_splatting 2026.2.9 is out.

• Faster ray tracing — reduced per-ray payload and improved stochastic transparency
• Updated sample projects
• Several portability and stability fixes

Full release notes:
https://nvpro-samples.github.io/vk_gaussian_splatting/


Meet LCC Scan, the Spatial Camera in Your Phone!

Scanning with iPhone is now as easy as shooting a video. With LCC Scan, you can capture people, objects, and small scenes as high-quality 3D spatial photos that you can freely explore.

Open LCC Scan, choose the right mode, and start capturing.
• As simple as 123, 1-min Scan, 2-min Wait, 3D Spatial photo is done!
• Free and easy to use, with everything generated locally on your phone—no cloud uploads and no server queues during peak times
• Share 3D orbit videos and interactive 3D models with one tap
• Export 3DGS models from LCC Scan via LCC Cloud as PLY and LCC2 format

LCC Scan is now available on the App Store. Download it today and create your first 3D spatial photo.

Download: https://apple.co/4xPl1RX
Learn more: https://bit.ly/4cgg6Re
Export from LCC Scan:https://www.youtube.com/watch?v=ltSptDcUk3M

An Android version is planned for a future release. Stay tuned


👤 Splat Avatars
Turn a Gaussian Splat scan of yourself into a fully animated avatar by selecting Splat under + New Avatar.
Follow the guided two-step workflow: first align your scan with the rig, then animate it.
The avatar automatically adjusts to changing lights and environments in real time, so it fits naturally into your Space.
Bring your own animations from Mixamo or any compatible source and apply them to your avatar.
-> Add Splat Avatar
📷 Image to Splat (splat.js)
Train a Gaussian Splat entirely in your browser without using a server, CUDA, or uploading your photos. Your data never leaves your device.
Simply drop in a set of photos, and Splat.js handles the camera solve, Gaussian initialization, and training directly in your browser tab.

-> Start Generation in your Browser
🤖 Guest Avatars
Give your visitors a custom default avatar and replace the standard default avatar. Available with PRO.
Use the Avatar Selection UI to choose outfits and dress your guest visitors before they enter the experience.
-> Define the syle for your Guests
✨ Enhanced Splat Agent
Meet the Space Agent, your personal assistant for building and managing experiences in Arrival.Space.
It can work across multiple Spaces, helping you manage projects and content beyond a single environment.
It can answer questions about how things work, help you Vibe Code, and place or configure objects directly in your Space.
Whether you need guidance, want to create something new, or want to make changes to your environment, the Space Agent is there to help.
-> Start working with the Agent
💻 Arrival.Space CLI
Set up your Space with the Arrival CLI by asking your favorite AI agent to build and configure your scene.
Use it to upload and place entities, add Vibes, manage assets, and prepare reusable Space Templates with predefined colors, layouts, and configuration values.
Before every push, a server-side snapshot is saved automatically, so changes can be validated and safely undone in the app if needed.
-> CLI Documentation




https://github.com/awesome-4dgs/awesome-4dgs

This list uses 4DGS broadly: native 4D Gaussian primitives, deformable or time-conditioned 3D Gaussians, explicit Gaussian trajectories or sequences, and closely related dynamic Gaussian representations. Of 351 listed arXiv papers, 329 are direct dynamic-Gaussian works, 2 are supporting resources, and 20 are clearly labeled adjacent or historical references.

Each category starts with a compact paper list: the title links to arXiv, followed by optional [page] and [code] links. Expand Detailed contributions for reviewed contribution summaries, tags, and repository-status icons. Focused sections intentionally repeat papers when that makes a section useful on its own.


https://www.fab.com/listings/44fcdaad-8765-4479-b2cb-adf34f9c0a4d
Bring Gaussian splats into Unreal Engine 🚀

WallGS is a native Gaussian Splat Renderer for Unreal Engine, built to make large splat scenes easier to import, render, optimize, and use inside real projects.

Import your existing .SOG or .PLY Gaussian splat files directly into the Content Browser, drag them into your level, and experience them in the Unreal viewport and at runtime on Windows, PCVR, Android, and standalone VR headsets such as Meta Quest.


Репост из: Physical AI
We built Splatimation mostly because we were frustrated.

We used to build tours on a platform designed around 360° panoramas. Gaussian Splats were bolted on as an afterthought.

The UI was clunky and importantly there was no streaming capability, so when you sent a tour to someone with less than a flagship device, it would stutter and lag.

That never sat right with us. 3D Gaussian Splatting is a better medium for virtual tours than panoramas ever were. Real depth, real parallax, real presence.

The one thing missing was a movement system that was built for the format instead of inherited from photo-sphere tech.

So we built one. That gap is what Splatimation fills. (and we think it's much easier to use too!)

Then we thought, wouldn't it be cool if it could move as well...

None of this would run in a browser the way it does without PlayCanvas.

Their decision to open-source their engine gave us the foundation to build Splatimation. So I want to say a massive thank you to them for enabling Splatreprenuers like us to create new platforms for the industry.

Source: https://www.linkedin.com/posts/harry-watts-79269098_we-built-splatimation-mostly-because-we-were-ugcPost-7483817137191604224-wR7C


Great news—we’re extending the presale until July 25th! ⏳

To be honest, we’ve been getting so many questions about the setup and compatibility that we wanted to give everyone enough time to make sure it’s exactly what they need for their gear before the price goes up.

If you’ve ever been out in the field shooting 360 or doing photogrammetry, you probably know the pain of a shaky rig. There’s nothing worse than opening the footage at work and realizing a heavy setup caused tiny vibrations that ruined the stitching or the 3D model. We really wanted to fix this, so we built the bracket to be rock-solid, even when you load it up.

And speaking of loading it up—mounting lights on a 360 rig without them showing up in the shot is usually a total nightmare. With this bracket, you can just snap the Halo 360 right above or below the lenses (or even use two lights at the same time for full coverage) and they stay completely invisible in the blind spot. No messy rigging, no hours spent fixing it in post-production.

So if you’ve been on the fence or waiting to ask us a question, you now have until July 25th to grab yours at the presale price.

Link is in the bio or comment, and if you’re unsure about anything, just drop us a message!

https://bushman-panoramic.com/products/bushman-bracket-35


Репост из: Physical AI
3D for everyone, everywhere: today, we’re proud to introduce OnTheFly.

https://onthefly3d.com

Born from the GraphDeco research group at Inria, OnTheFly is building the next generation of 3D media tools by putting high-quality 3D capture directly in the palm of your hand. Our ambition is to make capturing, navigating, and editing 3D content accessible to anyone, with no specialized hardware, no expert workflow, and no post-processing delay.

Photography was not always effortless. It once required technical expertise, careful preparation, and long processing times. Today, anyone can take a photo on a smartphone and see the result instantly. 3D is still in its film era.
Professional 3D capture often still requires dedicated hardware, skilled operators, long processing times, and a leap of faith: you only know whether the capture worked once the reconstruction is complete.

OnTheFly makes high-quality 3D capture intuitive, immediate, and accessible on any smartphone.

As you move through a space, the 3D reconstruction appears directly on your screen. You can see what has been captured, identify what is missing, and know immediately that you have the result you need. No blind capture. No post-processing stage. What you capture is what you see on the screen.

Founded by Anthony Schoofs, Andréas Meuleman, Camille Montemagni, and George Drettakis, OnTheFly builds on years of research in 3D reconstruction and novel-view synthesis. Our goal is to bring advanced 3D technology out of the lab and into real-world workflows across construction, manufacturing, marketing, and entertainment.

Our latest research will be presented at #SIGGRAPH2026, marking an important step toward immediate, robust, and scalable 3D reconstruction. We’ll also be exhibiting at the Cap Digital Pavillon France.

Come meet us and see what the future of accessible 3D looks like.

Source: https://www.linkedin.com/posts/siggraph2026-ugcPost-7483319158542155776-smnq


https://junyuandeng.github.io/Glob3r/
Global Structure-from-Motion with 3D foundation models, turning feed-forward geometric predictions into optimizable multi-view constraints


How to View Massive Gaussian Splats in the Browser.

This is a large Lublin, Poland scan with 250M+ Gaussian splats.
The raw PLY in my test was 16.4 GB — too large to import directly as a normal browser asset.

So instead of trying to open the raw file directly:
Convert first. View later.

The workflow:
PLY → Web-ready Package → TimeSplat 4D Business

After conversion, the scan loads quickly, streams detail progressively, and can be navigated smoothly in the browser. That preparation step is what makes very large splats easier to view, share and present.

https://splitview.studio/

https://www.youtube.com/watch?v=aihJ3RZVvpE


Видео недоступно для предпросмотра
Смотреть в Telegram
Jaskirat Singh

Today I’m releasing Splatline v2.

Splatline is a toolkit for exploring Gaussian splat videos in 3D. This release adds 5 reconstruction backends: VGGT, LongSplat, DepthSplat, SHARP, and TripoSplat.

It also includes faster pose and depth estimation, a tiered human pipeline, true splat rendering in the player, SLAM 3D mapping, image-to-mesh up to 11M vertices, and local Mac support.

The first release introduced the base system. v2.0 brings more of the full pipeline into one place.

#OpenCV #ComputerVision #3DReconstruction #GaussianSplatting #SLAM #OpenSource

https://github.com/jaskirat1616/Splatline


Видео недоступно для предпросмотра
Смотреть в Telegram
Rough 3D blockout → AI-photoreal still → a metric, camera-movable 3D Gaussian Splat. 🎬

That's a previz pipeline I've been building — and I just open-sourced the piece that makes the last arrow actually work.

The steps:

1️⃣ 3D scene (Houdini / Blender / etc.) — block it out, render a beauty pass + a camera-space depth AOV (Karma cam_zdepth / hitPz). The depth is real, to-scale.

2️⃣ Image model (an img2img pass, e.g. Nano Banana 2 / Gemini Flash Image) — turn the beauty into a photoreal RGB. Now it reads like a plate, not a blockout.

3️⃣ SHARP (this extension) — feed that photoreal RGB + the original metric depth into Apple's ml-sharp → out comes a 3D Gaussian Splat, at the correct real-world scale.

Step 3 is the bit I released: "Sharp Depth Injection."

On its own, ml-sharp turns one image into a splat — but it infers depth from a single view, so the result lands at an arbitrary scale. Nudge the camera and the geometry drifts. By injecting your real metric depth at inference time, the splat comes out to scale, so a few degrees of orbit or a metre of dolly hold together in 3D. Exactly the bar previz needs.

A few things others might find handy:

• It's a runtime add-on, not a fork — Apple's ml-sharp is never edited; a small subclass hot-swaps in at inference, so upstream updates still pull cleanly.

• A tiny HTTP service (image + depth EXR → .ply) you can drop into a comp tool, a web app, or a render pipeline.

• The trick that made it click: it grades SHARP's OWN predicted depth to match your reference, instead of forcing depth in pixel-by-pixel (which fights the model and throws "flying gaussians" at object edges). Clean geometry, correct scale.

If you live anywhere near previz, virtual production, or 3D-from-images, this whole rough-3D → AI → metric-splat loop might be worth a look:

https://github.com/gitcapoom/Sharp_Depth_Injection


https://www.linkedin.com/posts/simon-dewey-4b835622_single-video-source-4d-face-reconstruction-ugcPost-7481469343411703808-UKaa/?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAYuSGUB01rbMyTFNW4SkTf50dynCH7Luuw

😳Single video source 4D Face reconstruction to generative pipelines…

We’ve busy building away, exploring some interesting new ways to create…. Combining 4d face reconstructing from a single video input.

With automated character reference and motion transfer allowing for more control in motion transfer generative camera moves and recapturing performance from different angles. Very early days with this tech but the potential really is mind blowing. Beginning to get some really interesting scene based test working, keen to really stress test this beast out, let’s hope it holds up at scale🤞

…. Please ignore the sudden jump in performance, it was a naming error on the ply sequencing, all fixed now, but too tired to run out a new one 😅

Показано 20 последних публикаций.