How to Plot precision recall curve for semantic segmenttaion

I have done training and testing for semantic segmentation. I want to make a graph of my precision recall. I have got precision and recall values when validating images. but i don't know how to plot percurve charts. The following is the code that I made

pic = 8;
I = readimage(imdsTest, pic); %ib
Ib = readimage(pxdsTest,pic);
BC = labeloverlay(I, Ib,'Colormap',cmap,'Transparency',0.5);
C = semanticseg(I, net);
B = labeloverlay(I,C,'Colormap',cmap,'Transparency',0.5);%cb

imshowpair (B, BC,'montage')
pixelLabelColorbar(cmap, classes)
title ('prediction vs groundtruth')
expectedResult = readimage(pxdsTest,pic);
actual = uint8(C);
expected = uint8(expectedResult);
imshowpair(actual,expected)

iou = bfscore(C,expectedResult);
[recall,precision,prediction]=bfscore(C,expectedResult)
table(classes',recall ,precision, iou)

the results of this program are as follows

ans = 
        Var1        recall     precision      iou  
    ____________    _______    _________    _______

    "crack"         0.34446     0.25696     0.34446
    "background"    0.53501     0.60656     0.53501

to be able to make precision recall plots. What should I do? Please help me

[Ismat Mohd Sulaiman]

I'm studying this shared script. Hope it will help yours.

https://www.mathworks.com/matlabcentral/fileexchange/21528-precision-recall-and-roc-curves

Answer 1

[Chinmay Budhiraja]

Hi,

According to my understanding, you want to study the precision recall curve for a task. We use ROC curve (Receiver Operating Characteristic Curve) for the same. Consider the following example:

load simplecluster_dataset
net = patternnet(20);
net = train(net,simpleclusterInputs,simpleclusterTargets);
simpleclusterOutputs = sim(net,simpleclusterInputs);
plotroc(simpleclusterTargets,simpleclusterOutputs)

In semantic segmentation, generally a multi-dimensional image is converted into a vector first by flattening (you can use reshape for the same) the matrix and then roc is plotted. You can tweak the above sample according to your use case. Please refer to plotroc for more information.