search

google-ai-edge / mediapipe

Cross-platform, customizable ML solutions for live and streaming media.
https://ai.google.dev/edge/mediapipe
Apache License 2.0
27.44k stars 5.15k forks source link

Facemesh for mobile devices #508

Closed tttzof351 closed 4 years ago

tttzof351 commented 4 years ago

Yesterday google released new facemesh model as part of mediapipe, but for mobile devices it is not available. Will it be added later?

https://blog.tensorflow.org/2020/03/face-and-hand-tracking-in-browser-with-mediapipe-and-tensorflowjs.html

These packages are also available as part of MediaPipe, a library for building multimodal perception pipelines:

mgyong commented 4 years ago

@tttzof351 We released a version of the face mesh for TF.js. What's your use case for the mobile version?

We are looking to release this. Timeline is undecided

bhack commented 4 years ago

@mgyong I have a more general question. Is there any planned activity in the roadmap to enable more cross platform examples instead of having to reimpl for TPU, JS, Mobile and Desktop?

mgyong commented 4 years ago

@bhack Most of our main examples like Multi hand tracking, object detection is cross platform: mobile (iOS, android), desktop (C++). TPU(Coral) & TF.js face mesh/handpose examples are not cross platform due to implementation reasons and are the exception. For example, TPU(Coral) requires quantized model for Coral EdgeTPU and the TF.js face mesh/handpose examples are pure js for pre-processing and post processing.

Our general plan has always been to enable MediaPipe for cross platform on-device inference pipelines