i have a problem with modelGradients

7 vues (au cours des 30 derniers jours)
Hind Haboubi
Hind Haboubi le 4 Mai 2021
Commenté : debojit sharma le 16 Juin 2023
When i run thus section :
for numEpoch = 1:nEpochs
reset(preprocessedTrainingData);% Reset datastore.
iteration = 0;
while hasdata(preprocessedTrainingData)
t_start = tic;
% Custom training loop.
% Read batch of data and create batch of images and
% ground truths.
outDataTable = read(preprocessedTrainingData);
XTrain = outDataTable{1,1}{1};
YTrain = outDataTable{1,2}{1};
if isempty(YTrain)
continue;
end
% Convert mini-batch of data to dlarray.
XTrain = dlarray(single(XTrain),'SSCB');
% Evaluate the model gradients and loss using dlfeval and the
% modelGradients function.
[gradients,boxLoss,objLoss,clsLoss,totalLoss,state] = dlfeval(@modelGradients, model, XTrain, YTrain,yoloLayerNumber);
I get this error :
'modelGradients' is used in Generate Synthetic Signals Using Conditional Generative Adversarial Network.
Error in deep.internal.dlfeval (line 18)
[varargout{1:nout}] = fun(x{:});
Error in dlfeval (line 41)
[varargout{1:nout}] = deep.internal.dlfeval(fun,varargin{:});
Error in nouveux (line 94)
[gradients,boxLoss,objLoss,clsLoss,totalLoss,state] = dlfeval(@modelGradients, model, XTrain,
YTrain,yoloLayerNumber);

Réponse acceptée

Arianna Pryor
Arianna Pryor le 12 Mai 2021
I got a similar error when trying to run the VAE example. I had to use another modelGradients function.
function [infGrad, genGrad] = modelGradients(encoderNet, decoderNet, x)
[z, zMean, zLogvar] = sampling(encoderNet, x);
xPred = sigmoid(forward(decoderNet, z));
loss = ELBOloss(x, xPred, zMean, zLogvar);
[genGrad, infGrad] = dlgradient(loss, decoderNet.Learnables, ...
encoderNet.Learnables);
end
  1 commentaire
debojit sharma
debojit sharma le 16 Juin 2023
I am also facing the same problem while trying to train VAE for RGB image. @Arianna Pryor Can you please kindly tell me in which part of the code have you used this modelGradients function?

Connectez-vous pour commenter.

Plus de réponses (1)

Mohamed Marei
Mohamed Marei le 26 Juil 2021
Modifié(e) : Mohamed Marei le 26 Juil 2021
Because MATLAB sees both of these functions which share the same name on the path, it doesn't know which one to use for each example. Therefore, it may be better practice to copy the internals of this function into a different function (more appropriate for your example), i.e.
function [gradients, boxLoss, objLoss, clsLoss, totalLoss, state] = ...
yoloModelGradients(network, Xtrain, Ytrain, yoloLayerNumber)
% loss, gradients, and states definitions go here.
end
After that, replace the call to the old modelGradients function in your call to dlfeval:
% Evaluate the model gradients and loss using dlfeval and the
% yoloModelGradients function.
[gradients,boxLoss,objLoss,clsLoss,totalLoss,state] = dlfeval(@yoloModelGradients, model, XTrain, YTrain,yoloLayerNumber);
Hope you've managed to get it fixed since then!

Catégories

En savoir plus sur Deep Learning for Image Processing dans Help Center et File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by