Overview
With this release, we can support Linux ARM64 and AMD64. I dropped support of the face_recongition, which was responsible for face recognition over video frames, but I had a lot of issues getting it to work cross platforms (arm specially) this issue.
I switched to using deepface, which was easier to migrate, and I don't need to precompile dlib for the project use case and the model that I'm using (VGG-Face). And reduce the Docker build time and image size. More details here
And now, the face labeling will be folder-based, and we don't need to load face encoding into a json file. I can pass the faces folder to deepface and they handle the rest. More details here
And I dropped the emotion detection plugin, and now I'm using face recognition to do both in one shot (face and emotion recognized). More details here