 Hello, and welcome to Arrow Electronics Tech Snack. Today, we are going to present you a basic AI demo running on the Avenger 96 board. This development board is hosting the first MPU from ST Microelectronics. The STM32 MP1 is combining dual ARM Cortex A7 with an ARM Cortex M4 core. While the A7 cores are giving the performance to run Linus-based systems with rich graphical user interface and advanced networking ability, the M4 core can provide the real-time functionality for any application. In today's demo, we are attaching additionally a CAM1 mezzanine from Shiratec Solution. This is providing us the video data decoding to YUV format thanks to ONSEMI's 13MP image sensor and the corresponding image signal processor on the mezzanine itself. We are going to use this video stream as input for AI model. We are using the image segmentation model that is available on the Tensor4Light model page. This model function is similar to the standard object recognition, but the output will be the segment or area of the picture that contains the recognized object instead of the probability value of the identified object class. So let's have a closer look on the demo itself. First, let's get through the data flow again in details. The camera module is based on a 13MP image sensor from ONSEMI. It provides the video data in raw format for having it an easily processable format. We are converting the stream into YUV with ISP AP1302 from ONSEMI. At this point, our data is still transferred on MIPI lines that are not available on the host processor. So we use the STMIPID02 MIPIDESALIRIZER to change our interface to parallel. When the processor receives the stream, it rescales it for the AI model requirements and then combines the interference output with the original video input. As last step, the processor outputs the final result on an HDMI monitor. In our demo, we are using the output information of the model to apply a well-known fake background effect on the video stream. As an additional feature, you can change the virtual background by simply clicking on the change background button. Further improvement on the AI algorithm can be done by quantization and transfer learning methods. If you are interested in the details, please visit the product page of Avenger 96 or the CAM1 mezzanine at arrow.com to find more information and the complete guide for the demo. Please follow the Arrow Tech Snack next episodes to see other cool and interesting videos on the newest Devkitsen demos.