Ultralytics

Driver fatigue can turn a normal journey into a serious safety risk in seconds. 🚗 Detecting the warning signs early is one of the most practical applications of computer vision in transportation. "cty-tnhh-gii-php-cng-ngh-rlab" is helping move that forward. The driver's drowsiness dataset is now on the Ultralytics Platform: 9,467 images. 9,467 annotations. 2 classes: drowsy and awake. That focused data provides a strong foundation for developing real-time driver monitoring systems that can detect changes in driver alertness and support timely warnings. From commercial fleets and logistics to public transportation and passenger vehicles, this dataset can help developers explore computer vision systems designed to make journeys safer. Thank you, "cty-tnhh-gii-php-cng-ngh-rlab", for contributing this dataset and sharing it with the Ultralytics community. 🙏 Building computer vision for safer roads? 👉 Explore the dataset on the Platform ➡️ https://lnkd.in/ewaF_zsQ

New tutorial | Build a client-server computer vision system with Ultralytics YOLO26 🌐 What happens when computer vision needs to operate across multiple cameras, devices, or edge systems instead of a single machine? In this tutorial, we build a peer client-server workflow with YOLO26 and Python, where clients perform inference and send their results back to a central server for processing, monitoring, or downstream analysis. What'll be covered: ✅ Peer client Python implementation ✅ Peer server Python implementation ✅ Starting and configuring the server ✅ Running YOLO26 inference on client device ✅ Sending inference results from clients to the server This architecture can support multi-camera monitoring, industrial systems, smart facilities, retail analytics, and other deployments where multiple edge systems need to share Vision AI results with centralized infrastructure. Watch now ➡️ https://lnkd.in/euVSgTV7 #Ultralytics #YOLO26 #SocketProgramming #VisionAI

Build, train, and explore computer vision models in one interactive workspace. 🚀 JupyterLab keeps your entire workflow in one place: write and run experiments interactively, visualize predictions and metrics inline, and document findings alongside code-without switching between tools. Get started: ✅ Install JupyterLab and the Ultralytics Python package ✅ Launch JupyterLab ✅ Open the Ultralytics tutorial notebook ✅ Experiment with Ultralytics YOLO26 locally Explore the integration ➡️ https://lnkd.in/eaaTmzmC #Ultralytics #YOLO26 #JupyterLab #ComputerVision

New release Ultralytics v8.4.160 | Dataset edits now refresh caches 🔄 Editing labels in place now triggers a dataset rescan, helping keep training data current. Try the update and share your feedback! Minor updates: ✅ Lower peak VRAM for SAM3 text prompting ✅ Training retries preserve optimizer state after out-of-memory errors ✅ Validation reuses the checkpoint’s dataset when available Ultralytics v8.4.160 release notes ➡️ Read the release notes https://lnkd.in/eK7eBu5Y

Safer checkpoint loading, faster cold starts, and better compatibility. ⚡ Ultralytics v8.4.150 makes restricted checkpoint loading more targeted while restoring support for fused YOLOE and older checkpoints. Instead of scanning every supported model module, the update resolves only the model classes referenced by a checkpoint, reducing unnecessary work while maintaining trusted namespaces and allow-list protections. It also improves privacy by preventing Sentry from capturing command-line arguments and local variables that may contain credentials: small changes, but meaningful improvements for safer and more predictable production workflows. Read the release notes ➡️ https://lnkd.in/e-UJnmUC #Ultralytics #MachineLearning #MLOps

New tutorial | Package segmentation with Ultralytics YOLO26 📦 In this tutorial, we train a YOLO26 segmentation model on the package segmentation dataset using the Ultralytics Platform, evaluate its performance, deploy it to an endpoint, and run inference locally through the deployed model. What'll be covered: ✅ Dataset overview ✅ Configuring and training YOLO26 on the Platform ✅ Understanding training metrics ✅ Testing trained model predictions ✅ Deploying the trained model to an endpoint ✅ Running local inference through the deployed endpoint From parcel handling and warehouse automation to logistics analytics and automated inspection, this workflow demonstrates how a custom segmentation model can move from dataset and training to deployment and application integration. Watch now ➡️ https://lnkd.in/enn45HRE #Ultralytics #YOLO26 #WarehouseAutomation #Logistics

Cutting inference costs by 28% with Ultralytics YOLO26. RTM builds industrial safety monitoring that runs on manufacturing sites' existing cameras, with no new hardware required. Every camera leverages Ultralytics YOLO26 models, feeding a deterministic event layer that catches falls and hazard-zone intrusions in real time. The results: ✅ 28% lower per-frame inference cost after moving to TensorRT and FP16 ✅ Zero lines of training code changed when migrating to YOLO26 ✅21 manufacturing sites live or under evaluation Read the full case study: https://bit.ly/4cUZrTs

Embedded World North America 🚀 Ultralytics will be featured at the STMicroelectronics booth at embedded world Exhibition&Conference North America in Anaheim from September 22 - 24. Meet the STMicroelectronics team to learn how we are bringing real-time Ultralytics YOLO inference to STM32N6 microcontrollers. Learn more ➡️ https://bit.ly/46IciF4

Monitor heavy equipment across large industrial sites with computer vision! 🏗️ From cranes and trucks to pumps and bulldozers, keeping track of equipment across complex worksites can be challenging. Ultralytics YOLO26 can detect and track multiple equipment types from aerial footage, providing better visibility into site activity. This can support equipment monitoring, asset utilization, and operational oversight across construction, energy, and other large-scale industrial environments. Read more ➡️ https://lnkd.in/eb3f3HGJ #Ultralytics #YOLO26 #IndustrialAutomation #ComputerVision

New release Ultralytics v8.4.159 | Faster, lighter training and export workflows 🚀 Smarter split handling avoids unnecessary downloads, startup metrics improve visibility, and INT8 SavedModel exports use up to 18.2% less peak memory. Minor updates: ✅ More predictable validation and test split selection ✅ Immediate system metrics when training starts ✅ Lower memory usage during INT8 SavedModel export Ultralytics v8.4.159 release notes ➡️ View the release https://lnkd.in/esSG5hxy