Order of sampling negative examples

asked 2017-04-12 10:07:30 -0600

Lets say I have many images for negative sampling listed in the background file. But the resolution of these images is so high (720 by 1280) compared to the size of sliding window (15 by 21) that even one image quantitatively contains enough negative examples for training the whole cascade. Does this mean traincascade will sample all negative examples from the picture which is listed first in the background file?

edit retag flag offensive close merge delete

add a comment

1 answer

Sort by » oldest newest most voted

answered 2017-04-14 07:42:06 -0600

StevenPuttemans

20029 ●16 ●82 ●207 http://stevenputtemans...

Yep it does...

Basically what OpenCV does is

Open image 1, apply sliding window negative sample gathering for the image, without overlap.
Go to the next image, and reapply sliding window until images are finished.
IF the list is depleted, apply a single pixel padding, and reinitialize the process.
Continue until padding is impossible.