Open Sballest opened 13 years ago
Dear Sballest:
First Issue: This is already being done. If you look at the tldExample.m code, the positions of the bounding boxes across all of the frames are being stored in tld.bb. tld.bb is a 4 x N array, where in each column, the first two elements denote the x,y co-ordinates of the top-left corner, and the last two of the bottom-right corner.
Second Issue: These are stored as pX, pEx, nX, nEx. pX and pEx. pX and pEx are the feature vectors, and image patches of object model over the video sequence that are labelled positive, and are belonging to the object. nX and nEx are the same as the previous, but are for the negative labelled ones. All you would have to do is modify the code so that you skip the training, and re-use these before starting to tracker code. I haven't done it myself, but that's what you'd do.
Hope this helps,
Oh yes... just to make it clear, for the first issue, N is the total # of frames in the image sequence... just so that there's no confusion.
Good luck!
Hey there!
I would like to use TLD for my PhD and I have two issues that maybe somebody could help me with.
First, I Need to record the position of the box at each frame and write it in my output data. Any idea?
Second, I would like to make TLD able to re-use what it has learned in a previous tracking. Any suggestion?
All the best!