How to do activation quantization of EACH LAYER in lgraph or dlnetwork?

for activation of each layer in dlnetwork or lgraph, how to do quantization?

My dlnetwork is just a piece of network, not a whole network till the end.

My dlnetworks contains 3 layers of Conv2d, no classification layer.

I only use it to do inference.

So no training-aware quantization is needed.

Only post training quantization is needed.

Answer 1

[Matt J]

Have you seen this already?

https://www.mathworks.com/help/deeplearning/quantization.html

[Runcong Kuang]

Thanks. I thought this is for weight, bias. I just saw it also applies to activation. Good. But I failed to use this function because of some compiler problem even if I use 'matlab' as environment. So I have to manually quantize weights and biases to put into each layers, but can't do the same with activations.