CUDA_ERROR_OUT_OF_MEMORY

i am training 318 images of 1024 1024 1 size. These are the properties of my GPU.

         Name: 'Quadro K6000'
                     Index: 1
         ComputeCapability: '3.5'
            SupportsDouble: 1
             DriverVersion: 9
            ToolkitVersion: 8
        MaxThreadsPerBlock: 1024
          MaxShmemPerBlock: 49152
        MaxThreadBlockSize: [1024 1024 64]
               MaxGridSize: [2.1475e+09 65535 65535]
                 SIMDWidth: 32
               TotalMemory: 1.2885e+10
       MultiprocessorCount: 15
              ClockRateKHz: 901500
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 1
          CanMapHostMemory: 1
           DeviceSupported: 1
            DeviceSelected: 1

 I am using minibatchsize '5'.

layers = [

    imageInputLayer([1024 1024 1]);
    convolution2dLayer(3,16)
    batchNormalizationLayer;
    reluLayer();
    averagePooling2dLayer(2,'Stride',2);
    dropoutLayer
    convolution2dLayer(3,32);
    batchNormalizationLayer;
    reluLayer();
    averagePooling2dLayer(2,'Stride',2);
    dropoutLayer
    fullyConnectedLayer(2);
    softmaxLayer();
    classificationLayer()];

I get CUDA out of memory error.Help please.

[Joss Knight]

I'm a little surprised by this, although it is unusual to be using such high resolution images at the input. Try setting the stride of the first convolutional layer to 2 or 3 to get the resolution down quicker. Use the Network Analyzer to ensure that you are down to 1x1 in spatial dimensions by the time you get to your final Fully Connected layer (you are on 254x254, which means your network is unlikely to be very effective anyway).

[Saira charan]

I've changed the stride to 3 but it doesnot make any difference. I'm using MATLAB R2017b and network analyzer is for R2018a. What would you suggest, i don't want to lose much information, what size should i resize my images?

[Joss Knight]

All the standard networks use ImageNet data at 227x227 or 224x224. Can you upgrade MATLAB?