aihacker111 / Efficient-Live-Portrait

Fast running Live Portrait with TensorRT and ONNX models
MIT License
120 stars 10 forks source link

[Feature Request] support mediapipe key points? #21

Open salahzoubi opened 1 month ago

salahzoubi commented 1 month ago

Hello,

I'd like to ask if it's possible that you include mediapipe face extractor for landmarks in addition to insight face. Mediapipe seems to be a super popular framework for kp extraction and would love to see it in this awesome repo!

This repo already has it:

https://github.com/kijai/ComfyUI-LivePortraitKJ/tree/main

aihacker111 commented 1 month ago

@salahzoubi it can be done , but I’m noted another face landmark, because mediapipe is cannot run on GPU, so I need to rethinking and choose another model

aihacker111 commented 1 month ago

@salahzoubi InsightFace is use only for research and study , so i’ll update more option , will have mediapipe for you can choose. Stay tuned, now i’m already prepare to update IP-Adapter to it

salahzoubi commented 1 month ago

I believe mediapipe is extremely light weight and fast to run even without gpu. Especially while streaming realtime...

aihacker111 commented 1 month ago

@salahzoubi Thank you for your good suggestion , so i’ll update mediapipe for this. Stay tuned my friend