The Future of Object Detection support

As far as i understand it:

  • the tf1 OD API was deprecated in a half finished state (because the TF2 OD API was developed)
  • the tf2 OD API was deprecated in a half finished state (because the Model Maker was developed)
  • the Model Maker was abandoned in development (because the Media Pipe is getting developed)

So, when will the Media Pipe be abandoned? It is currently in v 0.1, what are the chances this project makes it to v1.0?

Is there an actual chance that Media Pipe will be actually developed for a longer time?
I am asking because i am responsible for implementing training pipelines for tflite models (OD tasks) and it seems like a more future proof approach would be to implement my own ObjectDetector.

How relevant is it even for the TF-Team to implement and support OD-Tasks for TFLite?

2 Likes

Hi @Malte_Kleine,

Apologies for the delayed response. If you are still seeking guidance, we are happy to inform you that the MediaPipe Model Maker is continuously improving. The latest version, 0.2.1.4, is now available. Additionally, we plan to introduce further Task API customization features in the future. Currently, we support customization for seven APIs, which you can explore here.

We recommend exploring the MediaPipe Model Maker, and should you encounter any issues or need support, please feel free to raise them in the MediaPipe GitHub repository. We will review your queries and provide assistance as needed.

Thank you!!