Week 13 progress

This week I resumed integrating the CI job for the inference model environment. I encountered some issues while trying to get the CI jobs checks successfully in the GitHub workflow environment. Currently, the CI job workflow is represented in a pull request, which should be merged as soon as I address the reviewer’s requested changes.

Workflow caches

The caches were not working for both of gRPC and OpenCV built dependencies, resulting in inefficient check runs. The main cause was that CMAKE_STAGING_PREFIX CMake argument used when building the libraries did not match the declared cache path.

System dependencies

Another issue with missing admin permissions when trying to make install the zlib1g-dev package after building it from source. This issue was resolved by installing the zlib1g-dev as a system library at the beginning of the workflow. In contrast, I switched the spdlog library to be a submodule instead of installing it as a system package. This approach is followed as spdlog is not needed to be a system library, which in turn mitigates the risk of breaking other packages.

Matching Yocto versions

I started to match the versions of used libraries (OpenVINO, OpenCV and gRPC) with the available versions in the master branch of Yocto. I checked each library version on Yocto branches using OpenEmbedded Layer Index

Next steps

  1. Complete and merge the current CI job.
  2. Improving the current inference models and automate dataset generation.
  3. Fix CMake issues in the inference model.