Toradex shows their deep learning inference solutions at Embedded World 2019, using power-efficient, Arm-based System on Modules, as machine learning and deep learning using neural networks progress is accelerating with successful new applications in computer vision for the embedded world. Toradex simplifies the integration of these technologies into products with its System on Modules, training deep learning models on high-performance computers with frameworks like TensorFlow, with optimizations needed to improve performance on low-power embedded Linux devices such as the ones Toradex partners with Au-Zone, Xnor.ai and Antmicro and others to bring to the embedded market. Implementing these dedicated neural network accelerators can boost the performance of embedded devices while keeping power consumption low, as shown in Toradex’s solutions with Intel Movidius Myriad and Gryfalcon Lightspeeur. Toradex partners with Allied Vision to showcase the brand-new Alvium industrial MIPI CSI-2 camera as a crucial component in the pasta detection demo which uses an Apalis System on Module featuring the NXP i.MX 8 QuadMax SoC with Cortex-A72, Cortex-A53 and dual OpenCL-capable GPU. To learn more about the demo, see CNX-Software’s blog post. A real-world application example of deep learning is Manta, a camera-based drowning detection system from Coral Detection Systems which is solar-powered, and the video analytics are done on a Toradex Apalis module featuring a Nvidia TK1 SoC with a CUDA-enabled GPU. If a person is at risk of drowning, the system can alert its user acoustically or via smartphone.