How to generate Custom Bitstream for Zedboard to deploy Neural Network model?

imds = imageDatastore('Result_fish_images(NA)', ...

'IncludeSubfolders',true, ...

'LabelSource','foldernames');

%%

[imdsTrain,imdsValidation,imdsTest] = splitEachLabel(imds,0.7,0.15,0.15,"randomized");

%%

numTrainImages = numel(imdsTrain.Labels);

idx = randperm(numTrainImages,16);

figure

for i = 1:16

subplot(4,4,i)

I = readimage(imdsTrain,idx(i));

imshow(I)

end

%%

classNames = categories(imdsTrain.Labels);

numClasses = numel(classNames)

%%

net = imagePretrainedNetwork("alexnet",NumClasses=numClasses);

net = setLearnRateFactor(net,"fc8/Weights",20);

net = setLearnRateFactor(net,"fc8/Bias",20);

%%

inputSize = net.Layers(1).InputSize

%%

pixelRange = [-30 30];

imageAugmenter = imageDataAugmenter( ...

'RandXReflection',true, ...

'RandXTranslation',pixelRange, ...

'RandYTranslation',pixelRange);

augimdsTrain = augmentedImageDatastore(inputSize(1:2),imdsTrain, ...

'DataAugmentation',imageAugmenter);

%%

augimdsValidation = augmentedImageDatastore(inputSize(1:2),imdsValidation);

%%

options = trainingOptions("sgdm", ...

MiniBatchSize=10, ...

MaxEpochs=6, ...

Metrics="accuracy", ...

InitialLearnRate=1e-4, ...

Shuffle="every-epoch", ...

ValidationData=augimdsValidation, ...

ValidationFrequency=3, ...

Verbose=false, ...

Plots="training-progress");

%%

net = trainnet(augimdsTrain,net,"crossentropy",options);

%%

scores = minibatchpredict(net,augimdsValidation);

YPred = scores2label(scores,classNames);

%%

idx = randperm(numel(imdsValidation.Files),4);

figure

for i = 1:4

subplot(2,2,i)

I = readimage(imdsValidation,idx(i));

imshow(I)

label = YPred(idx(i));

title(string(label));

end

%%

YValidation = imdsValidation.Labels;

accuracy = mean(YPred == YValidation)

%%After the above we performed the quantization and saved the network in quantizedNet variable. We flashed the %%memory card with linux image for zedboard using SoC Blockset support package. We tested the communication between %%our zedboard and laptop via zynq() command and were able to retreive the IP Address of it. Now we want to deploy the %%trained model on zedboard platform using:

%%hTarget = dlhdl.Target('Xilinx','Interface','Ethernet');

%%hW = dlhdl.Workflow('Network',quantizedNet,'Bitstream','zcu102_int8','Target',hTarget);

%%dn=hW.compile;

%%hW.deploy;

%%output=hW.predict(InputImg);

%%This should give us prediction result by performing the operation on FPGA and fetching back the result.

%%But here the pre built bit streams like zc0102 or zc706 are not available for zedboard. How to generate custom %%bitstream targetting zedboard ??

Answer 1

[praguna manvi]

Hi,

To generate custom bitstream for your target please refer to the documentation below which illustrates on how to configure custom bitstream using “dlhdl.ProcessorConfig;” object :

https://www.mathworks.com/help/deep-learning-hdl/ug/generate-custom-bitstream-to-meet-custom-deep-learning-network-requirements.html

Additionally, here is an example for deploying on an FPGA, specifically the Xilinx ZC706 SoC, with bitstream customization:

https://www.mathworks.com/help/deep-learning-hdl/ug/customize-bitstream-to-meet-resource-requirements.html#responsive_offcanvas

[A Lalith]

Hey thanks for replying.

We have trained the AlexNet for underwater images already and we want to deploy it on Zedboard (or Zc702 if possible because we have only these 2 boards in our university labs). We tried generating custom bitstream for it using below commands:

%%

hPC = dlhdl.ProcessorConfig;

%%

hdlsetuptoolpath('ToolName','Xilinx Vivado','ToolPath','D:\Vivado\2023.1\bin\vivado.bat');

%%

dlhdl.buildProcessor(hPC);

It runs till the end and tells that Workflow is complete but it doesnt generate bitstream file like in this below link:

https://in.mathworks.com/help/deep-learning-hdl/ug/generate-custom-bitstream.html

dlprocessor.bit isnt being generated. Rest all files are created.

Any other method we can try getting the bit file for trained neural network model in matlab??