Computer Vision at the Edge: Training and Deploying Ultralytics YOLO Models Across Real Hardware

Details

Through hands-on Python code, participants will train Ultralytics YOLO models, including the latest YOLO26, across a comprehensive set of computer vision tasks: object detection, instance segmentation, pose estimation, oriented object detection (OBB), image classification, multi-object tracking, and open vocabulary detection. Open vocabulary models extend traditional detection beyond fixed category lists, enabling flexible recognition driven by natural language, a powerful paradigm for real-world deployment scenarios.

A significant portion of the workshop is dedicated to the full edge deployment pipeline, and we will not shy away from its complexity. Compiling a model for edge hardware is rarely a straightforward process. It involves understanding the specific constraints of each target platform, navigating toolchains that differ significantly across vendors, managing quantization precision and its impact on accuracy, handling operator compatibility issues, and validating that the compiled model behaves as expected on device. We will walk through this process step by step, demystifying quantization, compilation, and hardware-specific model export so that attendees come away with a realistic and practical understanding of what production edge deployment actually involves.

The workshop goes beyond slides and code. Attendees will have access to physical edge hardware devices brought directly to the session, as well as live online connections to additional platforms, allowing participants to witness real inference performance across multiple chips firsthand. This includes platforms such as STMicroelectronics STM32N6, Intel OpenVINO, DeepX, Axelera, NVIDIA Jetson, and many more. Rather than simply showing benchmark numbers, we will demonstrate how these results are achieved and how attendees can replicate the same process on their own hardware targets using the Ultralytics ecosystem.

In addition, the workshop will feature contributions from hardware and software partners operating at the frontier of edge AI. These partners will share insights into the engineering work behind their platforms, covering topics such as NPU architecture, vendor-specific SDK toolchains, model optimization strategies tailored to their hardware, and the practical lessons learned from deploying vision models at scale on constrained devices. This gives attendees a rare opportunity to engage directly with the teams building the chips and tools that power modern edge AI systems. Participants will leave with working code, hands-on deployment experience across multiple real hardware platforms, and a solid and honest framework for navigating the full complexity of bringing Ultralytics YOLO models from training to the edge in their own projects.

TARGET AUDIENCE

Machine learning engineers, computer vision practitioners, embedded systems developers, and hardware engineers interested in production edge AI deployment across diverse platforms.

PREREQUISITES

Basic familiarity with Python and deep learning concepts.

No prior experience with Ultralytics or edge deployment required.

Visit organiser website