For object detection, the ground truth bounding boxes are imported as text instead of array from an excel or text file.

For deep learning based object detection, I have stored the ground truth in an excel (or txt) file. The file names of images are stored in one column and the bounding boxes in the format [x1,y1,width1,height1;....] for each image are stored in another column. When read using readtable function, the bounding boxes are also read as text. How should these be read as Mx4 arrays?

[Walter Roberson]

Is it correct that the [ and ] and comma and semi-colon are all part of the format?

Answer 1

[Walter Roberson]

filename = 'AppropriateName.txt';
S = fileread(filename);
info = regexp(S, '(?<name>\S+')\s+\[(?<bb>[^]]+)\]', 'names');
bboxes = cellfun(@(s) reshape(str2double(s),4,[]).', regexp({info.bb}, '[,;]', 'split'),'uniform', 0);