UTIA has published (for public access) two new application notes and evaluation packages on facedetect demonstrator within the StorAIge project:
- Xilinx Vitis AI facedetect Demo on Trenz Electronic TE0820 4EV SoM with TE0701 06 Carrier Board and Avnet HDMI In/Out FMC Card
- Xilinx Vitis AI facedetect and resnet50 Demo on Trenz Electronic TE0802 02 with ZU2CG and 1 GB LPDD4
First demonstrator demonstrates interface to Full HD HDMI video input/output and also interfacing to built in hardened video decoder.
It provides a tutorial describing setup and run Vitis AI 2.0 facedetect demonstrator on Trenz TE0820-4EV SoM with Trenz TE0701-06 carrier board and Avnet HDMI In/Out FMC card. The system uses a Xilinx DPU unit to accelerate calculations. An input video data are taken from the HDMI input (FMC), processed data are displayed via HDMI output (FMC). The system desktop is shown via HDMI output on TE0701 carrier board. As the TE0820-4EV provides H.264/H.265 Video Codec Unit (VCU), the demonstrator is configured to use the decoder of the VCU.
Second demonstrator demonstrates HW accelerated AI inference on reduced size DPU suitable for small device and low cost board. It provides a tutorial describing setup and run Vitis AI 2.0 facedetect and resnet50 demos on Trenz TE0802-02 board with ZU2CG and 1GB LPDD4. The system uses a Xilinx DPU unit to accelerate calculations. The DPU unit used in zcu104 or zcu102 does not fit into the used ZU2CG device.
For this reason, a reduced DPU configuration is provided as well as precompiled models for demos. This text also describes how to move the system filesystem to the M.2 PCIe SSD.
These two demonstrators will be presented by Lukas Kohout and Zdenek Pohl (both from UTIA) at the Embedded World 14-16/03/2023 at the Trenz Electronic booth: Hale 2, Stand 2-209.