CUDA Error - Semantic Segmentation

Hello,

Using semantic segmentation, I want to separate the satellite image into two classes, water and land. I am having this problem: "An unexpected error occurred during CUDA execution. The CUDA error was: CUDA_ERROR_LAUNCH_FAILED"

How can I solve this problem ?

Here my codes :

 clear;clc;close all

 dataDir = fullfile('C:\Users\firat\Desktop\TEZ\Uygulama\Semantic Segmentation\data');
 imDir = fullfile(dataDir,'image');
 pxDir = fullfile(dataDir,'imagePixelLabels');

 imds = imageDatastore(imDir);

 I = readimage(imds,1);
 figure
 imshow(I)

 % imageLabeler(imDir);

 classNames = ["Water" "Land"];
 pixelLabelID = [1 2];
 pxds = pixelLabelDatastore(pxDir,classNames,pixelLabelID);

 C = readimage(pxds,1);

 B = labeloverlay(I,C);
 figure
 imshow(B)

 buildingMask = C == 'Water';
 figure
 imshowpair(I, buildingMask,'montage')

 % Create a Semantic Segmentation Network

 numFilters = 64;
 filterSize = 3;
 numClasses = 2;
 layers = [
    imageInputLayer([1024 1024 3])
    convolution2dLayer(filterSize,numFilters,'Padding',1)
    reluLayer()
    maxPooling2dLayer(2,'Stride',2)
    convolution2dLayer(filterSize,numFilters,'Padding',1)
    reluLayer()
    transposedConv2dLayer(4,numFilters,'Stride',2,'Cropping',1);
    convolution2dLayer(1,numClasses);
    softmaxLayer()
    pixelClassificationLayer()
    ]

 opts = trainingOptions('sgdm', ...
    'InitialLearnRate', 1e-3, ...
    'MaxEpochs', 100, ...
    'MiniBatchSize', 64);

 trainingData = pixelLabelImageSource(imds,pxds);

 net = trainNetwork(trainingData,layers,opts);

 testImage = imread('C:\Users\firat\Desktop\TEZ\Uygulama\Semantic Segmentation\test\test3.tif');

 C = semanticseg(testImage,net);
 B = labeloverlay(testImage,C);
 figure
 imshow(B)

[Sean de Wolski]

What's the output from?

>>gpuDevice

[Firat Erdem]

ans =

CUDADevice with properties:

                      Name: 'GeForce 840M'
                     Index: 1.00
         ComputeCapability: '5.0'
            SupportsDouble: 1
             DriverVersion: 9.00
            ToolkitVersion: 8.00
        MaxThreadsPerBlock: 1024.00
          MaxShmemPerBlock: 49152.00
        MaxThreadBlockSize: [1024.00 1024.00 64.00]
               MaxGridSize: [2147483647.00 65535.00 65535.00]
                 SIMDWidth: 32.00
               TotalMemory: 2147483648.00
       MultiprocessorCount: 3.00
              ClockRateKHz: 1124000.00
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 1
          CanMapHostMemory: 1
           DeviceSupported: 1
            DeviceSelected: 1

Answer 1

[Joss Knight]

This is almost guaranteed to be due to a kernel time-out - your GPU is also driving your graphics, and Windows imposes a time-out on long-running kernels to prevent the graphics freezing up. Try setting the TdrLevel registry key to 0 to turn off the time-outs, and see if the problem goes away.