Reference Designs

Reference Designs#

Vitis AI provides two reference designs for the VEK385 evaluation board. Both demonstrate a complete, end-to-end pipeline on the Versal AI Edge Series Gen 2 platform, but they target different use cases:

Reference Design

Input Source

Output

VEK385 AI Reference Design

Image or video files

Inference results saved to disk

VEK385 MIPI Streaming AI Reference Design

Four live 1080p MIPI camera streams

Composited 4K DisplayPort output with inference overlays

Select the file-based design to evaluate accelerated pre-processing and NPU inference on stored media. Select the MIPI streaming design to evaluate a live, multi-camera capture-to-display system.

Common Requirements#

Both designs require the setup described in System Requirements, and both are delivered in the release package extracted per Software Setup.

VEK385 AI Reference Design#

This reference design integrates hardware-accelerated pre-processing with machine learning inference on the Neural Processing Unit (NPU). It demonstrates a complete, end-to-end pipeline, from ingesting image or video inputs to producing and saving the inference results.

Supported Features#

  • Hardware-accelerated pre-processing via the Image Processing PL kernel, including resizing, color space conversion, normalization, and cropping.

  • Inference execution on the NPU.

  • Software-based post-processing on the CPU.

  • Visualization utilities that overlay bounding boxes and labels (text) on the input image based on inference results.

Reference Design Contents#

  • Source code for the System Integrated Reference Design.

  • Sample applications that demonstrate ML inference accelerated on the AI Engines, hardware-accelerated pre-processing using the image_processing PL HLS (High-Level Synthesis) kernel, and CPU-based post-processing.

  • Ready-to-test boot images that embed the application binary for immediate evaluation on target hardware.

Download the Reference Design#

Ensure the reference design has been extracted per the instructions in Software Setup.

  1. VEK385 AI Reference Design source code (Rev-A)

  2. VEK385 AI Reference Design source code (Rev-B)

  3. Prebuilt boot images (Rev-A): vai6.3_vek385_RevA_ai_ref_design_boot_images.tar

  4. Prebuilt boot images (Rev-B): vai6.3_vek385_RevB_ai_ref_design_boot_images.tar

  5. SDK to cross-compile applications on the host machine: vai6.3_vek385_sdk.sh

The release package includes:

  • Boot images

  • ONNX model

  • VART-based and ONNX Runtime-based applications

  • Source code for the VAIML Platform Reference Design

Deploying the Pre-built Reference Design#

AMD provides the pre-built reference design image, along with a collection of C++ example applications, that you can run directly on the board to evaluate performance. These applications let you exercise the reference design without building anything from source.

For step-by-step instructions on running the C++ applications with the pre-built reference design image, see Reference Applications.

Building from Source#

To build the VEK385 AI Reference Design from source yourself, see (Build the VEK385 AI Reference Design).

VEK385 MIPI Streaming AI Reference Design#

This reference design is an end-to-end, live video implementation on the Versal AI Edge Series Gen 2 platform that combines multi-camera video capture, AI inferencing, and 4K display output.

The design captures four live MIPI camera streams at 1080p through the Versal hardened Image Signal Processor (ISP), performs hardware-accelerated streaming pre-processing in the Programmable Logic (PL), runs AI inference on the AI Engine (AIE), and displays the composited results with inference overlays on a 4K DisplayPort monitor.

Supported Features#

  • Simultaneous capture of four live 1080p MIPI camera streams through the Versal hardened ISP.

  • Hardware-accelerated streaming pre-processing implemented in the PL.

  • AI inference execution on the AI Engine (AIE).

  • Four-stream composition and display on a 4K DisplayPort monitor, with inference results overlaid on the live video.

Hardware Requirements#

In addition to the common requirements, this design requires:

  • VEK385 Rev-B evaluation board. Rev-A boards are not supported for this design.

  • FMC-96716A daughter card.

  • Four supported MIPI camera sensors connected through the FMC-96716A daughter card. The following sensors are supported:

    • OX03F10: OmniVision 3MP, 12-bit RAW HDR sensor

    • Sony IMX728: 8MP, 12-bit RAW sensor

  • A 4K-capable DisplayPort monitor.

Download the Reference Design#

Ensure the reference design has been extracted per the instructions in Software Setup.

  1. VEK385 AI Reference Design source code (Rev-B)

  2. Prebuilt boot images (Rev-B): vai6.3_vek385_RevB_mipi_stream_ai_ref_design_boot_images.tar

Design documentation#

For complete build instructions, hardware setup, camera connection details, and the step-by-step procedure for running the design on the board, see the VEK385 MIPI Streaming AI Reference Design documentation.

Contents