Which landmark extraction model was used during training?

[cvipym]

Thanks for great job!

I want to know landmark extraction model to implement wing loss.
Could you please tell me which landmark extraction model was used?

Did you convert the onnx model from InsightFace, as shown in the code, to a torch model?
Or did you use another model, such as Mediapipe, for landmark extraction?

[cleardusk]

We use the self-developed model: https://huggingface.co/KwaiVGI/LivePortrait/blob/main/liveportrait/landmark.onnx

[cvipym]

Thank you for reply :)