Loci

Greetings from Milan 🇮🇹 Last week, our ML Research team represented Loci at the European Conference on Computer Vision (ECCV). It is one of the world’s largest, most respected forums of its kind. Three thousand scientists from top universities and industrial labs—like NVIDIA, Meta, and Google—gathered two miles from the Duomo to present frontier research. Topics ranged from gaussian splats, to LiDAR segmentation, to video diffusion. At the Loci booth, we shared out latest work on teaching machines to perceive and describe 3D objects. A delightful week, with delightful people. To our fellow attendees, thank you for your fantastic demos and seminars. To the folks at the European Computer Vision Association, thank you for organizing a wonderful event. Pictured: Jack Davis (🔜 ECCV), Atharva Deshmukh, Jose Caballero, Akshay Gupta

Loci is hiring a ML Research Engineer to join us at our headquarters in London. 3D recognition is one of AI’s last, hardest frontiers. Working on that frontier means Loci gets to claim a lot of ‘mosts’ and ‘firsts’—like building the most accurate 3D labeling models in the world, and pioneering the very first 3D LLMs. Along the way, we do cutting-edge research in computer vision, machine learning, graphics, and language modeling. We come from deep research backgrounds at MIT, Oxford, Cambridge, Princeton, Imperial, Meta, and Twitter, and spend our days solving specific problems exceptionally well. It’s hard, frequently delightful work. And it can’t be done alone. The Loci ML Team is looking for curious and talented engineers in London to help us create the first, best foundational models for 3D. If that sounds like you—or someone you know—visit us at loci.ai/jobs

“Creating machines that understand the physical world requires teaching them [to] understand not just text and images, but also 3D objects,” writes Loci Machine Learning Scientist Akshay Gupta. Learn more about how Loci is teaching machines to understand 3D in our ML Research team's latest article, “Describing 3D objects using Natural Language.” Link in comments 👇

Our friends at echo3D have built a world-class product. Now, it’s powered by world-class AI. We’re thrilled to announce that Loci is partnering with echo3D to bring AI tags and natural language search to their 3D asset management platform. Today, 135,000 designers around the world use echo3D to store and share vast libraries of 3D assets in real time. Loci can now automatically generate descriptive tags for those assets—so they always show up in search and filters. The result? echo3D users can find the right asset in seconds, reuse work across projects, and never need to tag by hand. Because 3D is the future, and working with it should be effortless. We'll share a link to echo3D's announcement in the comments 👇

Last Wednesday, 120+ builders and investors came to the AWS London headquarters for AI Demo Days to discuss the future of AI, and the AI of the future. If reports are to be believed, it was the largest event in the space yet—and the fastest to sell out. Loci’s Head of Machine Learning, Jose Caballero, spoke to a packed room about our work with 3D vision language models. A fantastic evening with a fantastic audience. Thank you to our fellow presenters, to Amazon Web Services (AWS) for hosting, and to Matt Carey and the good people at StackOne for making it all possible. Photos by Adam Isfendiyar 📸

A secret ingredient. Loci's AI is currently the most accurate in the world at recognizing 3D assets. One method that sets us apart is that we “preprocess” assets to make them easier to recognize. We extract point clouds, generate 2D renders, and, when needed, swap file formats. All to make assets as parsable as possible. This summer, we added a powerful step to this flow: background plane removal. Planes are the large rectangles that sit below or around a 3D object. They can help with lighting and act as colorful backgrounds. But, for our purposes, they distract from the object itself. By removing them, we allow our AI to focus on what matters. The result? Even more accurate recognition. Plane and simple.

Meet Auto Caption. Describe complex 3D assets and scenes with a single API call. We piloted Auto Caption this summer on user-generated scans (powered by Niantic Lightship) of neighborhoods here in London. The results have been awesome. Why 3D scans? Because having a digital record of the physical world is not just cool, it’s critical. Scanning the real world helps engineers analyze construction sites for safety, video game designers create immersive environments, manufacturers evaluate machine parts, and so much more. And, thanks to apps like Lightship, Capturing Reality's Reality Scan, Luma AI, Autodesk ReCap, and Trimble Inc. Realworks, creating these scans can be as easy as pulling out your phone. Auto Caption adds the missing text data that makes scans and assets possible to search and understand in seconds. Special thanks to our very own William Giles for cooking this demo up. As we always say, “where there’s a Will, there’s a way.” Try it out for yourself. Link in comments 👇

What happens in Vegas? On November 11th, Loci is joining AI startups and titans like NVIDIA and Amazon Web Services (AWS) at the 2024 Trimble Inc. Dimensions conference. We’re honored to be in such great company. Dimensions is one of the largest conferences in the world for physical industries like architecture, construction, engineering—and, this year, AI. Because AI is more than chatbots and sparkle emojis. It can be profoundly subtle and gritty—doing hard work behind the scenes to make human life measurably better. At Dimensions, Loci will unveil a project we’ve been working on as a SketchUp 0-60 Challenge finalist. We’ll explore how architects can use computer vision for 3D scenes to make buildings safer, more sustainable, and easier to design. Subtle, gritty, game changing. We can’t wait to show you. Learn more by joining us in Vegas or online. Details in the comments 👇

At Loci, we spend our days turning frontier research into practical software. Tonight, at 6:30 PM in London, we get to explain how. Our friends at Amazon Web Services (AWS) and StackOne are sponsoring a talk about how Loci built one of the world's first 3D LLMs—and our plans to make it better. Come meet our Head of Machine Learning, Jose Caballero, who will be giving the presentation, and our CTO, Jack Davis, who you can identify as the plant in the crowd clapping the hardest. Details below 👇

📣💂 London-based ML researchers & developers: a quick plug for a Loci technical talk on September 4 at 6:30pm. Our friends at AWS and StackOne are sponsoring a talk about how we built our own 3D LLM and our R&D roadmap to make it better. Come meet our Head of Machine Learning Jose Caballero, who will be giving the presentation, and our CTO Jack Davis, who you can identify as the plant in the crowd clapping the hardest. 👏 Register here to join: