HOW TO CALCULATE RECALL, PRECISION AND IoU test data deep learning

Dear all,

I want to calculate precision and recall for my test data. But I gor Error. Because my data is 3D. (as attached)

[precision,recall] = bboxPrecisionRecall(volMask1,tempSeg1)

ERROR

Error using bboxPrecisionRecall
Expected boundingBoxes to be two-dimensional.

Error in bboxPrecisionRecall>validateNonTableInput (line 153)
validateattributes(bbox, {'numeric'},...
    
Error in bboxPrecisionRecall (line 110)
validateNonTableInput(boundingBoxes, 'boundingBoxes');

Answer 1

[Anusha]

Hi,

I understand that you are trying to calculate the precision, recall and IoU metrics on the deep learning predicted output and groundtruth. I also see from the .mat files attached that your volumetric groundtruth (volMask1) and predicted output (tempSeg1) are of the size 128x128x64.

The bboxPrecisionRecall() function currently supports only 2-D inputs for bboxes and groundTruthBboxes. Therefore, convert the 3-D volumes into 2-D images and you can refer to the following code that does this:

% Access 2-D images from 3-D volume and find the metric average

avgPrecision=0; totPrecision=0;
avgRecall=0;totRecall=0;

for i= 1:size(volmask1,3)
    [precision,recall] = bboxPrecisionRecall(volMask1(:,:,i),tempSeg1(:,:,i));
    totPrecision=totPrecision+precision
    totRecall=totRecall+recall
end

avgPrecision = totalPrecision/size(volmask1,3);
avgRecall = totalRecall/size(volmask1,3);

Please refer to the following documentation for more details regarding precision recall computation on the data:

Compute bounding box precision and recall against ground truth - MATLAB bboxPrecisionRecall (mathworks.com)

Thanks,

Anusha

[mohd akmal masud]

but I got problems

Error using bboxPrecisionRecall
Expected boundingBoxes to be of size Mx4, but it is of size 128x128.

Error in bboxPrecisionRecall>validateNonTableInput (line 153)
validateattributes(bbox, {'numeric'},...
    
Error in bboxPrecisionRecall (line 110)
validateNonTableInput(boundingBoxes, 'boundingBoxes');