Real-Time Vision AI on Embedded Artists RZ/G3E with MemryX and Virtium
This demonstration illustrates how the combination of Embedded Artists hardware, Virtium's industrial M.2 accelerator, and MemryX AI acceleration enables real-time vision applications on an embedded Linux platform.
MemryX recently demonstrated a complete embedded vision AI solution built on the Embedded Artists RZ/G3E System-on-Module (SoM) and carrier board, paired with a Virtium M.2 AI accelerator powered by MemryX technology.
The demonstration showcased real-time object detection from a live USB camera stream using a YOLO26 model trained on the COCO dataset. Running at a 640×640 input resolution, the application sustained the full 30 FPS camera rate while simultaneously performing inference and rendering results.
Rather than relying solely on the host processor for AI workloads, the MemryX accelerator architecture executes neural network inference while the RZ/G3E manages image processing, application logic, and display. This division of responsibilities enables modern computer vision applications on compact embedded platforms without the need for a discrete GPU or high-power compute module.
End-to-End Embedded Vision
The demonstration consisted of:
- Embedded Artists RZ/G3E Developer's Kit
- Virtium VMX-004 M.2 AI accelerator powered by MemryX
- Single USB camera input
- YOLO26 object detection model
- 640 × 640 inference resolution
- Real-time GUI with annotated video output
- Sustained 30 FPS end-to-end performance
The application performed live object detection directly on the embedded platform, illustrating how AI acceleration can be integrated into existing industrial and edge computing systems while maintaining real-time responsiveness.
Open Software Integration
Support for the Embedded Artists platform is enabled through the recent MemryX Yocto integration for RZ/G3E-based systems, allowing developers to build complete BSP images with MemryX runtime and driver support included.
A core goal of the MemryX software ecosystem is to make AI acceleration straightforward to integrate into embedded Linux platforms. By providing open-source Yocto layers and maintaining support for standard embedded development workflows, developers can incorporate MemryX acceleration into their existing BSPs with minimal changes, rather than maintaining custom software stacks or proprietary build environments.
Developers can find the Embedded Artists integration in the official GitHub repositories:
embeddedartists
Yocto BSP layer for Embedded Artists Renesas platforms
embeddedartists
MemryX integration layer for Embedded Artists platforms
The MemryX Yocto layers can be added directly to the BSP build, simplifying deployment on RZ/G3E-based hardware while providing a reusable foundation for future AI-enabled embedded designs. This modular approach makes it straightforward to extend MemryX support across additional Yocto-based platforms, enabling developers to accelerate new hardware without reinventing their software infrastructure.
Expanding Embedded AI Capabilities
Many embedded systems already provide excellent CPU performance for control logic, networking, industrial protocols, and user interfaces. As AI workloads continue to grow, dedicated neural accelerators provide an efficient way to add high-performance inference without fundamentally changing the underlying platform architecture.
