Which face detection model was used for the v1.5 model and v1.6 models?

[paulgeorge]

My assumption is that face_alignment model was used for training and the v1.5 checkpoint release.

However, I'm not sure if the v1.6 model checkpoints continued to use face-alignment or used the insightface model (or is it a mix of both?)

If anyone knows for sure I'd be most grateful for some clarification on this.