Question about vl_dspcovsift.m

[jizhihang]

Hi, I have a question about the vl_dspcovsift.m . In the source code
`[~, assign] = sort(f_size);

f_sort = f(:, assign);
[~, assign_back] = sort(assign);
[f_out,d] = extract(im,'sift',f_sort);

f_out = f_out(:, assign_back);
d = d(:, assign_back);`

Why do we need to sort the f and resort f_out back again? Why do we input the f into the function "extract" directly? Sorry for my poor English. Look forward to you.