How to access the INT8 quantized weights in the deep learning quantizer?

I have quantized Resnet 18 using the deeplearning quantizer toolbox. The idea is that I want to deploy it on an FPGA. The quantized process was successful and the model size was compressed to 10MB. However, I want to see the quantized INT8 weights. How do do I access it in the terminal? I can only see the floating point values.

Below is my code and the output in the terminal:

code

% Save the network temporarily to calculate size

save('quantizedNet.mat', 'quantizedNet');

fileInfo = dir('quantizedNet.mat');

netSizeMB = fileInfo.bytes / (1024^2);

fprintf('Quantized Network Size: %.2f MB\n', netSizeMB);

%% Network archieture view

%%analyzeNetwork(quantizedNet)

%% Quantizer details

% Choose layer and parameter

Layers=qDetails.QuantizedLearnables

conv1weight=Layers.QuantizedLearnables.Value(1)

conv1weight

output

Answer 1

[Dor Rubin]

Hi Yousef,

You can access the integer representation by using the storedInteger method on the fi value. For example:

storedInteger(conv1weight)

Thanks,

Dor