Roboflow

T-4 days until GTC! 🚀 📍 Meet Seeed team at Booth #1703 Looking back at last year’s GTC, it’s incredible to see how far we’ve come. Now, with just four days to go, we can’t wait to connect with partners, customers, and the amazing developer community! Beyond our booth, don’t miss these exciting sessions and demos we’re co-creating with our partners: ✅ Amazon Web Services (AWS) Workshop: "Unleash Generative AI Microservices at the Edge With NVIDIA Jetson and AWS IoT Greengrass" [DLIT74570] 👉 Sign up here: https://lnkd.in/gyX29RYM 👀See Seeed Jetson Orin devices @ AWS Booth #715 ✅ Roboflow @ Booth #344 Explore reComputer, pre-built with Roboflow Inference—enabling seamless inference pipelines, a visual workflow editor, and access to state-of-the-art models like Florence-2, CLIP, and SAM2. 💡 On the weekend of GTC, join our Embodied AI Hackathon & Meetup! In collaboration with Hackster.io and Circuit Launch, this hands-on hackathon will bring together developers to build 3D-printed robot arms using Hugging Face’s LeRobot and NVIDIA Robotics Jetson Orin, exploring imitation learning and Diffusion Policy/ACT methods. 👉 Register now: https://lu.ma/vnndgrnf 🦾 Hackathon: March 22–23, 2025 ⭐️ Demo Day & Meetup: 5PM, March 23, 2025 Can’t wait to see you all at GTC! Let’s innovate together. 🚀 #GTC2024 #AI #EdgeAI #GenerativeAI #JetsonOrin #AIRobotics

Day 3 at #EW2025! Meet Seeed #reCamera & reCamera Gimbal at Embedded World 2025! reCamera is your shortest pathway to building a market-ready AI camera!  We are at 📍 Stand 3-556 ⭐️reCamera + Ultralytics YOLO 11n: Discover how the reCamera counts people and works with a Local LLM controller to predict and intelligently control lighting and ventilation systems. ⭐️reCamera Gimbal Experience Zone: Get hands-on with our console to control the gimbal and switch between models (Roboflow-built, ONNX, and others) to track custom targets defined by you! 🎮 Come and play with them!

📏 Live session: Let's learn how to measure the size of objects using vision AI and then refine accuracy with a few techniques, like camera calibration and distortion correction. Register here: https://lu.ma/sliprclx

Super fun project. We looked at data from thousands of custom AI models, built for enterprise usage, and extracted a few cool and useful nuggets of information. Read the blog: https://lnkd.in/g8aA6KrC Download the report: https://lnkd.in/g2M3x8md My favorite parts: 💡 43% of enterprise models were trained with datasets containing fewer than 1,000 images. 💡 One organization retrained their AI model 18 times during the sample period, expanding the training dataset by over 900 images, and improved the accuracy 40 percentage points. 💡 A national food supplier is using a small form factor NVIDIA Jetson Nano detect packaging defects in real-time. #visionAI #computervision #roboflow

Previews and hints of what we are releasing soon in this podcast episode, check it out to hear the latest from our CEO Joseph Nelson

25,000 GitHub Stars ⭐ ⭐ ⭐ TL;DR: Supervision, the open-source library I started over 2 years ago, has now crossed 25,000 stars on GitHub! Thanks to everyone who contributed—especially Onuralp SEZER 👨🏻💻, Linas Kondrackis, and Hardik Dava! 🙏🏻 Over the past 2+ years, the project has grown to over 4,000 commits, more than 1,000 pull requests, and contributions from 100 open-source contributors. Supervision now powers nearly 5,000 open-source projects on GitHub, as well as almost every internal/external project at Roboflow. We're not slowing down! We're actively working on integrating Vision-Language Models (VLM) into traditional object detection pipelines and are gearing up for a massive object tracking integration in the near future. Stay tuned! 🔥 ⮑ 🔗 github repository: https://lnkd.in/d57M3c32 #computervision #opensource #programming #github

🚀 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗠𝗮𝗻𝘂𝗳𝗮𝗰𝘁𝘂𝗿𝗶𝗻𝗴 𝗤𝗖: Exploring automated defect detection systems 🤖 As a learning project to explore emerging Industry 4.0 technologies, I've been investigating how Computer Vision could drive tangible manufacturing improvements. Leveraging Roboflow's no-code platform and publicly available datasets (special thanks to the original authors), I developed an experimental workflow to detect defects in PCBs (Printed Circuit Boards). 𝗧𝗵𝗲 𝗩𝗮𝗹𝘂𝗲 𝗣𝗿𝗼𝗽𝗼𝘀𝗶𝘁𝗶𝗼𝗻 𝗳𝗼𝗿 𝗠𝗮𝗻𝘂𝗳𝗮𝗰𝘁𝘂𝗿𝗶𝗻𝗴 • 𝗖𝗼𝘀𝘁 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: Automated quality control reduces reliance on manual inspection • 𝗜𝗺𝗽𝗿𝗼𝘃𝗲𝗱 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆: ML-driven detection enables consistent, high-precision inspection • 𝗦𝗰𝗮𝗹𝗮𝗯𝗶𝗹𝗶𝘁𝘆: With appropriate training data, this approach adapts to various products • 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝗙𝗲𝗲𝗱𝗯𝗮𝗰𝗸: Immediate defect detection allows faster production adjustments • 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗹𝗲 𝗜𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻: Modern tools make Computer Vision practical for manufacturers of different sizes 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗝𝗼𝘂𝗿𝗻𝗲𝘆  𝟭. 𝗢𝗯𝗷𝗲𝗰𝘁 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗬𝗢𝗟𝗢𝘃𝟭𝟭: • Started with base dataset of 693 PCB images • Maximized training data through strategic augmentation (rotation, shearing) • Achieved 96.3% mAP, 97.4% precision, and 91.8% recall • Focused on common PCB defects including scratches and misaligned components 𝟮. 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗖𝗿𝗼𝗽𝗽𝗶𝗻𝗴 𝗳𝗼𝗿 𝗣𝗿𝗲𝗰𝗶𝘀𝗶𝗼𝗻: • Implemented region-of-interest focusing for improved defect localization • Enhanced detection accuracy in critical areas 𝟯. 𝗠𝘂𝗹𝘁𝗶-𝗟𝗮𝗯𝗲𝗹 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗩𝗶𝗧-𝗕𝗮𝘀𝗲: • Leveraged Vision Transformer architecture • Expanded dataset through augmentation to 2250 training images • Achieved 96.1% validation accuracy • Enables identification of multiple defect types 𝟰. 𝗘𝗻𝗱-𝘁𝗼-𝗘𝗻𝗱 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻: • Created unified detection, classification, and visualization pipeline • Developed streamlined workflow for defect analysis This personal exploration project helped me better understand how Industry 4.0 technologies can enhance manufacturing processes. Thanks to platforms like Roboflow that democratize computer vision capabilities, it's becoming increasingly feasible to implement ML-driven quality control solutions that deliver tangible business value. 𝗥𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀: • https://lnkd.in/gfzKwdZJ • https://lnkd.in/gzPuJYtV #computerVision #MLOps #manufacturingTech #learningInPublic #industry40