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
27.34k stars 5.13k forks source link

Fine-tuning face detection and face landmark models #5671

Open AbdullahOx opened 2 weeks ago

AbdullahOx commented 2 weeks ago

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

None

OS Platform and Distribution

Fedora Linux 40

Python Version

3.12

MediaPipe Model Maker version

No response

Task name (e.g. Image classification, Gesture recognition etc.)

Face detection and face landmark

Describe the actual behavior

na

Describe the expected behaviour

na

Standalone code/steps you may have used to try to get what you need

Hi, 

I have an unpublished dataset that I am using for research. Both face detection and landmark models will be useful.

I have tested the face detection model on my dataset and obtained an AP@0.5 score of 0.89.

I would like to improve these results by fine-tuning the model. I haven't yet tested the face landmark model, but I assume I will need to fine-tune it as well.

Thank you.

Other info / Complete Logs

No response

kuaashish commented 2 weeks ago

Hi @AbdullahOx,

Thank you for suggesting the extension of the model maker to include face detection and face landmarking features. Currently, these features are not available, and we do not have any documentation or code examples to reference. However, we can mark this as a feature request and share it with our team. Implementation will depend on future demand and bandwidth, but we can not provide a timeline at this moment.

AbdullahOx commented 2 weeks ago

Hi @kuaashish

Thank you for your reply. I'm looking forward to where this goes.