the function of trainMaskRCNN repot an error

hellow, everyon

I want to make a maskrcnn net for my data，

first I use Image Labeler

Second ,I make a dataStore，reference supporting documents

https://www.mathworks.com/help/vision/ug/getting-started-with-mask-r-cnn-for-instance-segmentation.html

[imdsRGB,pxds]=pixelLabelTrainingData(gTruth);
[~,blds] = objectDetectorTrainingData(gTruth);
imdsMask=imageDatastore("PixelLabelData_2\", ...
    "ReadFcn",@denseMasks)
imdsMask.Files=pxds.Files;
trainSet=combine(imdsRGB,blds,imdsMask);
function data=denseMasks(imds)
pixelLabelIDs=[1 3 4];%get from gtruth
h=480;%rgbImagesize
w=640;%rgbImagesize
numarray=numel(pixelLabelIDs);
Masksarray=false(h,w,numarray);
a=zeros(h,w);
I=imread(imds);
    for i=1:numarray
        n=pixelLabelIDs(i);
        a(:)=n;
        m=a==I;
        Masksarray(:,:,i)=m;
    end
data=Masksarray;
end

I think the data set I built is correct

Then，I construct a network

classnames={'Wheel','shallowRut','deepRut'};
inputsize=[480,640,3];
anchorBoxes=estimateAnchorBoxes(blds,3)
network = maskrcnn("resnet50-coco",classnames,anchorBoxes,"InputSize",inputsize)

Then,I make trainoption and start to train

do_train_maskRCN=1;
if do_train_maskRCN
    options = trainingOptions("adam", ...
    InitialLearnRate=0.001, ...
    LearnRateSchedule="piecewise", ...
    LearnRateDropPeriod=1, ...
    LearnRateDropFactor=0.95, ...
    Plots="training-progress", ...
    MaxEpochs=10, ...
    MiniBatchSize=3, ...
    BatchNormalizationStatistics="moving", ...
    ResetInputNormalization=false, ...
    ExecutionEnvironment="gpu", ...
    VerboseFrequency=50);
    trainedDetector = trainMaskRCNN(trainSet,network,options);
end

The error is at the last code

I have consulted all the relevant support documentation, I don't know what the problem is, is there a problem with my data set Settings ？

Answer 1

[YuechenW]

Hi Yan,

I'm not a MaskRCNN expert and I don't use it frequently. But here are some of my observations:

From the documentation, the second column of the data is an MX4 matrix, where M is the number of objects. So in your case, I think M is 1, right? The fourth column is a HxWxM matrix where HxW is the height and width of the image and M is the number of objects. Therefore, the 4th column cell should hold a matrix with size 480x640x1.

Train Mask R-CNN network to perform instance segmentation - MATLAB trainMaskRCNN - MathWorks United Kingdom

See a dataset sample size in this example.

Perform Instance Segmentation Using Mask R-CNN - MATLAB & Simulink - MathWorks United Kingdom