Merge the output from regionprops 'BoundingBox' into the format required for groundTruth.

The output from regionprops 'BoundingBox' from an image with multiple ROI's is, by default a 'struct' (structured array) with a row for each bounding box or a table. groundTruth (and boxLabelData) want the bounding box data as a comma seperated list. I can't work out how to combine the regionprops 'BoundingBox' data into the format required.

Output from regionprops for 1 image with 95 ROI's and 95 BoundingBox (95 bounding boxes in the 1 image)

This is an screen snip from a different data set and is just for context

gtruth table contents (for images 13 - 22) with bounding box data for 1 or 2 bounding boxes per image.

I feel like this should be really easy but can't work it out. I'm just learning Matlab and obviously a long way to go.

Cheers

Answer 1

[Matt J]

Let's say you have bounding box data from calling regionprops on some number of image frames. Then, your loop to organize the box data and labels might look like below:

numFrames=5; %number of frames

for i=1:numFrames

 Frame=rand(10)>0.6; %randomly generated image frame    
    
 S=regionprops( Frame,'BoundingBox');

 bbox=vertcat(S.BoundingBox);

 N=size(bbox,1);

 Boxes{i,1}=bbox;
 Labels{i,1}=string(randi(100,N,1)); %randomly generated box labels

end

T=table(Boxes,Labels)

T = 5×2 table
       Boxes           Labels   
    ____________    ____________

    {4×4 double}    {4×1 string}
    {3×4 double}    {3×1 string}
    {3×4 double}    {3×1 string}
    {6×4 double}    {6×1 string}
    {5×4 double}    {5×1 string}

[Charles Muir]

Thanks, I was soooo close.

Answer 2

[Image Analyst]

https://www.mathworks.com/videos/deep-learning-for-computer-vision-120997.html?s_tid=srchtitle_regression%2520cnn%2520vehicle_19

https://www.mathworks.com/help/vision/ref/trainfasterrcnnobjectdetector.html?s_tid=srchtitle_regression%2520cnn%2520vehicle_6

[Charles Muir]

Thanks for the links, will help me with other stuff.