MATLAB Answers

Error using nnet.inter​nal.cnngpu​.convolveB​iasReluFor​ward2D

20 views (last 30 days)
I am doing image segmentation with function semanticseg and trained network.
Everything goes well until the last update of my computer. It was conducted by the university and I don't know what were changed. The change I can see is that I used to train the network on single CPU,now I can train it on single GPU and for my network, the training speed is about the same.
However, I used to use the network trained with function semanticseg and the image is 4000*6000. Now the big image cannot be processed while a smaller one still can be. And the error is shown.
The code is like this, the training of the net goes well without problem.
testImage = imread('QRed00065.JPG');
Img=testImage;
imshow(Img)
% Segment the test image and display the results.
C = semanticseg(Img,net);
B = labeloverlay(Img,C);
imshow(B)
Out of memory on device. To view more detail about available memory on the GPU, use 'gpuDevice()'. If the problem persists, reset the GPU by calling 'gpuDevice(1)'.
Z = nnet.internal.cnngpu.convolveBiasReluForward2D( ...
Z = this.ExecutionStrategy.forward( X, ...
outputActivations = thisLayer.predict(XForThisLayer);
YBatch = predictNetwork.activations({X}, layerIndex, layerOutputIndex);
Y = this.calculateActivations(X, layerID, 1, varargin{:});
Y = this.UnderlyingDAGNetwork.activationsSeries(X, layerID, varargin{:});
allScores = activations(net, X, params.PixelLayerID, ...
[Lroi, scores, allScores] = iClassifyImagePixels(Iroi, net, params);

  0 Comments

Sign in to comment.

More Answers (1)

Joss Knight
Joss Knight on 3 Aug 2019
If you want to go back to using your CPU, add the 'ExecutionEnvironment' 'cpu' to your call to semanticseg.
C = semanticseg(Img,net,'ExecutionEnvironment','cpu');

  1 Comment

Jingyang Zhao
Jingyang Zhao on 5 Aug 2019
Thank you for that.
I wonder how to find out the biggest size of image that my GPU can process with.

Sign in to comment.

Sign in to answer this question.

Products


Release

R2019a

Translated by