Does TF-MOT quantization will keep use tfLiteConverter?

[JuHyung-Son]

I just noticed that quantization guide in tf-mot is using tfLite Converter.
Then, will TF-MOT library will partially use TF-Lite?

[alanchiao]

Hi @jusonn

In general, TFMOT is working to make our tools backend-agnostic (meaning support TensorFlow, TFLite, and other ones also).

As you can see in this issue, we are working to make latency improvements for both TF and TFLite via pruning.

We will be launching a tool for quantization-aware training soon. TensorRT mentions its predecessor, contrib.quantize, in its docs.

[alanchiao]

For a while, the post-training quantization tools will be available only through TFLite. Again, we working to generalize it.

Closing and feel free to reopen and if you don't think your question is fully answered.