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
27.42k stars 5.15k forks source link

Is there a way to improve performance through some configuration parameters? #5300

Closed huanhunx closed 6 months ago

huanhunx commented 6 months ago

Can face landmarkers optimize performance through some parameters during initialization or detection? For example, reducing some detection points, like stopping predicting irises as mentioned in this issue: https://github.com/tensorflow/tfjs/issues/4108

kuaashish commented 6 months ago

Hi @huanhunx,

Currently, it seems that achieving this using certain parameters in our Task API may not be possible. However, we will verify this and update you regarding any further possibilities.

Thank you!!

schmidt-sebastian commented 6 months ago

Reducing the number of outputs does not significantly affect performance as this woule be done as a post-processing step. Please ensure you are using a GPU delegate and that you input image size is small.

google-ml-butler[bot] commented 6 months ago

Are you satisfied with the resolution of your issue? Yes No