imtransform (in arrayfun()) does not work when called with images of much lower dimensions about cvpr15-cfss HOT 3 OPEN

Hi Grigoris, thanks for your remind. For now I just do a simple modification at Line 25. Yes as mentioned a good solution is to use imwarp. Thus we need either to reconstruct a new 2Daffine object for each original tform, or refactor the whole toolbox. I would possibly refactor the toolbox to fit in the new version of Matlab in the future. Currently I think the modification can let the codes work. Thanks!:)

from cvpr15-cfss.

It could.

I am not sure though, about the image and the bounding box loosing their correspondence. Isn't that significant in the process?

from cvpr15-cfss.

Hi. Maybe I have misunderstood the problem.

My last comment assumes the problem to be like this: inputs are fine, but some of them would output a smaller transformed image (e.g. in the codes, size(output,1) is smaller than the win_size 250). Then the vl_hog might extract an out-of-scope location of the image (e.g. x up to 200 > size(output,1)) and results in the crash.

I myself came across this problem before. This problem is not dependent on the input face size.

However, if the problem is caused by small-size input, then according to my experience, it seems that I did not come across such problem before. In other words, it seems the original face size seems not relevant to bugs. If this is the case for your problem, could you please provide more details to reproduce the problem? The bounding box and image should not lose their correspondence.

from cvpr15-cfss.