Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.

Possible bug in WarpBilinearQuad

gol
Beginner
433 Views

Sorry if I don't have the time to describe this more, it may be a false report, I won't really be using that function -it was only a test.

The screenshot shows the result of WarpPerspective vs WarpBilinear with the same settings. When the dest quad is a rectangle with integer coordinates (on the pic it's basically the source bitmap just zoomed out), it doesn't look very correct. At all other settings, WarpBilinear draws correctly, just not with these specific settings. It's also if the source bitmap is big enough.
But if you have no clue what could produce this weird output in the original algo, then maybe it's a problem on my side. Just reporting.

It's in all IPP versions.

http://www.flstudio.com/gol/Warp.jpg

0 Kudos
1 Reply
gol
Beginner
433 Views

let's say

source rectangle=(0,0),(243,0),(243,73),(0,73)

dest=(-0.25004,0),(199.75010,0),(199.75011,59.87463),(-0.25004,59.87463)

(source rectangle is also the same as source bitmap)

So it's basically a rectangle to another with only zoom, no rotation or other deformation.

I think the prob is those zero's, it's a smoothed animation and it takes some time until it glitches, when the values tend towards zero. It may even be because of denormals, who knows.

0 Kudos
Reply