Getting error when using semanticseg function

10 ビュー (過去 30 日間)
Manusree Bhatter
Manusree Bhatter 2020 年 7 月 28 日
回答済み: Harsha Priya Daggubati 2020 年 7 月 31 日
I am trying to use the semanticseg function to segment an image of dimensions 256x232x192 with a trained DAGnetwork. However I keep getting an error saying this:
"Error using gpuArray/cat. Maximum variable size allowed on the device is exceeded.
Error in nnet.internal.cnn.layer.Concatenation/predict (line 54)
Z = cat(this.ConcatenationAxis, X{:});
Error in nnet.internal.cnn.DAGNetwork/activations (line 571)
outputActivations = thisLayer.predict(XForThisLayer);
Error in DAGNetwork/calculateActivations (line 86)
YBatch = predictNetwork.activations({X}, layerIndex, layerOutputIndex);
Error in DAGNetwork/activations (line 138)
Y = this.calculateActivations(X, layerIndex, layerOutputIndex, varargin{:});
Error in semanticseg>iClassifyImagePixels (line 447)
allScores = activations(net, X, name, ...
Error in semanticseg (line 248)
[Lroi, scores, allScores] = iClassifyImagePixels(Iroi, net, params);"
My gpu has 24 GB of RAM, so it seems extremely unlikely that I am running out of memory running this, considering that the image gpuarray I am inputting is much less than 24 GB. I am able to input patches of [64 64 64] and [128 128 128] of the image succesfully and segment it, but I need to be able to input the whole image into the function as well. Is there a bug within the semanticseg function? How can I input the whole image into the semanticseg function? Thank you in advance!
  2 件のコメント
Walter Roberson
Walter Roberson 2020 年 7 月 28 日
It is not uncommon to start running out of memory when you have an array about 1/3 of the maximum memory of the GPU, and becomes quite likely when you have an array that exceeds half of the maximum memory of the GPU.
Manusree Bhatter
Manusree Bhatter 2020 年 7 月 28 日
The array is less than 1 megabyte, it is nowhere near the maximum memory of the GPU. Could there be an issue with the semanticseg function itself?

サインインしてコメントする。

回答 (1 件)

Harsha Priya Daggubati
Harsha Priya Daggubati 2020 年 7 月 31 日

Community Treasure Hunt

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

Start Hunting!

Translated by