Antmicro’s Smart Vision devkit with Toradex Apalis iMX8 gets Movidius / UP AI support

Throughout the past few years, Antmicro has been involved with a wide array of AI-capable computing platforms - ranging from the NVIDIA Jetson GPGPU family, through Xilinx FPGA MPSoCs, all the way to Qualcomm’s Snapdragon and NXP’s latest and most capable heterogeneous SoC, the i.MX8.

Sometimes the compute capabilities of a given platform need to be dynamically adapted as time goes by and use cases change. Since it’s not always feasible to just redesign an entire device from scratch, we often see our customers looking for smart and efficient ways to get the most out of their current platform and/or ensure that the next-gen solution we’re designing for them is future-proof.

Another way to future-proof both new and old designs is to leverage a fresh and interesting trend of additional, AI-focused accelerators which provide optional compute capabilities where needed. The UP AI core, a spiritual successor of the earlier Movidius USB Compute Stick in a more robust mini PCI express package, is a good example. UP AI Core allows you to bolt on a small and power efficient AI accelerator to any solution equipped with a mini PCI Express slot, such as our Apalis dual-camera devkit which supports both the Toradex Apalis TK1 and the iMX8 SoMs.

We’ve enabled the UP AI core on the i.MX8 version of our Apalis-based Smart Vision kit, basing on our previous work around adding our OV5640 camera support for Apalis iMX8 before our EW 2018 demo with Toradex. With the necessary drivers included, the Apalis iMX8 can process the incoming camera stream, performing NN computation on the UP AI core while leaving the i.MX8 CPUs and GPU free for other tasks.

The UP AI core is based on the Intel Movidius Myriad 2 VPU (Vision Processing Unit), and is capable of tasks like Deep Neural Network-based Classification, Pose Estimation, 3D Depth maps, Visual Inertial Odometry as well as Gesture/Eye Tracking and Recognition. Thus, the UP AI core can be used to extend the built-in edge AI capabilities of the i.MX8 platform (see e.g. NXP’s i.MX8-based Neural Networks for Object Recognition demo) with additional capabilities - or, if you’re running an older, non-AI capable CPU, enable those in the first place.

Movidius Myriad 2 is compatible with the Android 8.1 Neural Network API for on-device inferencing, which means you can leverage Antmicro’s Android services to build a NN-capable Android-based device, and get us to design the appropriate neural net setup adapted to the capabilities of the platform for your use case.

If you’re interested to learn what are the options to boost your edge device with the extra processing power enabled by the UP AI core, or if you want to build your next product leveraging both UP AI and our interchangeable Apalis iMX8/TK1 Smart Vision kit, contact our sales team at contact@antmicro.com.