Gaussian Splatting

🔥 “𝐆𝐚𝐠𝐚”: 𝐆𝐫𝐨𝐮𝐩 𝐀𝐧𝐲 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧𝐬 𝐯𝐢𝐚 𝟑𝐃-𝐚𝐰𝐚𝐫𝐞 𝐌𝐞𝐦𝐨𝐫𝐲 𝐁𝐚𝐧𝐤! 🔥 𝐌𝐨𝐭𝐢𝐯𝐚𝐭𝐢𝐨𝐧: The challenge of achieving consistent 3D segmentation across multiple views is significant. The innovative teams from 𝐔𝐧𝐢𝐯𝐞𝐫𝐬𝐢𝐭𝐲 𝐨𝐟 𝐂𝐚𝐥𝐢𝐟𝐨𝐫𝐧𝐢𝐚, 𝐌𝐞𝐫𝐜𝐞𝐝, 𝐍𝐯𝐢𝐝𝐢𝐚, and 𝐆𝐨𝐨𝐠𝐥𝐞 have developed “𝐆𝐚𝐠𝐚”, a framework that reconstructs and segments open-world 3D scenes by leveraging inconsistent 2D masks predicted by zero-shot segmentation models. Gaga effectively associates object masks across diverse camera poses, ensuring precise mask label consistency. 👉𝐏𝐚𝐩𝐞𝐫 https://lnkd.in/gxbsmBeB 👉𝐏𝐫𝐨𝐣𝐞𝐜𝐭 https://www.gaga.gallery/ 👉𝐑𝐞𝐩𝐨 https://lnkd.in/gymEw3_2 𝐊𝐞𝐲 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬: ✅ 𝐌𝐮𝐥𝐭𝐢-𝐯𝐢𝐞𝐰 𝐂𝐨𝐧𝐬𝐢𝐬𝐭𝐞𝐧𝐜𝐲: Gaga employs a 3D-aware memory bank to ensure consistent mask labels across different views. ✅ 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧 𝐒𝐩𝐥𝐚𝐭𝐭𝐢𝐧𝐠: Utilizes Gaussian Splatting for high-quality 3D scene reconstruction and segmentation. ✅ 𝐅𝐥𝐞𝐱𝐢𝐛𝐥𝐞 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧: Accommodates 2D segmentation masks from diverse sources, enhancing its versatility. #Gaga #3DSegmentation #AI #Innovation #Technology #WELEARN.ai

Testing Gaussian Splatting for Cultural Heritage Documentation in VR. This time I tested the scan of St. Andreas Church in Düsseldorf on PC VR. For this test, I imported the .PLY from the XGRIDS LCC Studio software in the Gracia AI VR player app and recorded the session wirelessly on my Meta Quest 2 headset. Even though, a mono recording of this experience cannot do the original justice, you still get a solid impression of the space. Considering that the scan was completed in about 35 minutes, I don't think there is a faster way to create such a VR experience. What do you think? 🏛️This wonderful building is a great example of early Baroque architecture. It was consecrated in 1629 and is now home to the Dominican Convent here in Düsseldorf. For the higher quality animation which I created in After Effects, please see my previous post. https://lnkd.in/eivrZyDq 🙏Thanks to Father Christoph for the kind permission to scan the church and to my friend Colt Chen from XGRIDS for giving me the chance to capture this exceptional building. #CulturalHeritage #GaussianSplatting #VirtualReality #3dScanning #Düsseldorf

Hey! I wanted to share with you a tool that I've developed for my small #gaussiansplatting research as I couldn't find an alternative: https://lnkd.in/d4ncFzfM It's a real-time visualizer for a Gaussian Splatting scene (or an Image tensor). This means that you can, for example: - bind your Gaussian Splatting scene to the visualizer - start the visualizer (detached from the main Python thread) - optimize the Gaussian Splatting scene - see it changing in real-time The advantage it takes is that it doesn't concur with the Python GIL (Global Interpreter Lock) as the visualizer is started in native code; and it reads CUDA memory, therefore there's no need to download the tensors to host memory.

I played around a bit more with the #4DGS models and combined a dynamic Gaussian model with a static Gaussian environment. This can be done easily directly in #Postshot now that you can drag multiple radiance fields into it.

https://lnkd.in/eqMyp6Wh #gaussiansplatting #3DGS Gaussian Splatting Gaussian Splatting

⚙️ Convert `.ply` to `.splat` i've created a simple space to convert .ply gaussian splat files to .splat format https://lnkd.in/eeyiJbsB

🎶 Paint with love, paint with light 🎶 This is lit by hand in a few seconds in my pitch black studio using 176 cameras at the Xangle Studio, processed in #gaussiansplatting With Kim Henry

3/3 papers accepted at #ICLR2025! 🎉 1️⃣ 6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering https://lnkd.in/ddFiN2Tf 🗨️ We leverage directional information in 6D space and enhance color/opacity representations to seamlessly boost 3D Gaussian splatting. 🙋 This project was spearheaded by my brilliant colleague Zhongpai Gao. 2️⃣ 3D Vision-Language Gaussian Splatting https://lnkd.in/dtFdkhJ6 🗨️ We present a new method for open-vocabulary 3D scene understanding, better balancing visual and language modalities to alleviate overfitting. 🙋 Our intern Qucheng Peng passionately worked on this all summer. 3️⃣ Order-aware Interactive Segmentation https://lnkd.in/d_sNDHiR 🗨️ We propose a new paradigm in interactive segmentation, leveraging the relative depth between objects to better isolate target entities. 🙋 This work was driven by yet another skilled intern, Bin Wang. Really proud of our team!

Turn any 3D Gaussian Splatting scene into gelatin using GSOPs for SideFX Houdini. 😂 This is possible via GSOPs Coarse Meshing, releasing soon (GSOPs 2.1). So, what should we call this feature? Join us on Discord for more shenanigans: https://lnkd.in/gdY79_Se Learn more about GSOPs: https://lnkd.in/gfSErWYb #GSOPS #GaussianSplatting #3DGS #SideFX #Houdini #3Dreconstruction #viewsynthesis #radiancefields #VFX #motiondesign #computergraphics #3D