CVML CI Tests

[sreeram-11]

User Actions

Windows

• CMake installed and available on PATH
• Ryzen AI CVML Library downloaded from AMD Account Portal and extracted to C:\RyzenAI-Library
• OpenCV 4.11 install (https://opencv.org/releases)
• Ryzen AI NPU driver (Download from https://ryzenai.docs.amd.com/en/latest/inst.html)
• Visual Studio 2022 Desktop Development with C++ workload

Linux

• CMake installed and available on PATH
  • sudo apt install cmake
  • Verify: cmake --version
• Ryzen AI CVML Library downloaded from AMD Account portal and extracted to ~/RyzenAI-Library
• OpenCV 4.11 install (Follow instructions in dependencies/opencv.md)
• Ryzen AI NPU driver
• Vulkan SDK
  • Follow README instructions
  • Verify with vulkaninfo --summary and glslc –version
• Additional required packages mentioned in the README for Ubuntu 24.04

What CI Tests Validate

1. Ensure prerequisites are installed and present:
   • CMake
   • OPENCV_INSTALL_ROOT env var set
   • OpenCV_DIR env var set
   • Visual Studio 2022 Desktop Development with C++ workload installed (Windows only)
   • Vulkan SDK tools installed (Linux only)
2. Clone https://github.com/amd/RyzenAI-SW.git
3. Run git lfs pull so large CVML binaries are downloaded
4. Check that Linux CVML runtime files are real binaries, not Git LFS placeholder files after navigating into RyzenAI-SW/Ryzen-AI-CVML-Library
5. Build the CVML sample apps with CMake
6. Confirm these executables exist:
   • cvml-sample-face-detection
   • cvml-sample-depth-estimation
   • cvml-sample-face-mesh
7. Add CVML runtime libraries and XRT libraries to LD_LIBRARY_PATH
8. Use assets/sample_face.jpg from the playbook repo and run face detection with the default model
9. Run face detection again with -m precise
10. Confirm output images were created and are not empty

IMP INFO

OpenCV 4.11 installation:

• Windows: Install using the official OpenCV releases page: https://opencv.org/releases
• Linux: Installation instructions have been added to dependencies/opencv.md.

[danielholanda]

@sreeram-11 Please take a look at #236 as well so we can merge that PR and ensure tests still look good after those changes.

[danielholanda]

Tests look good. Added @anna-amd-com for visibility.