Zedboard Warp Transform Reference Design

I am using the warp Transform reference design from UltraZed 3EG starter kit: https://avnet.egnyte.com/dl/FxDY3zWXMQ

When I changed TRANSFORM_TYPE from 0 to 1, which runs perspective transform instead of affine transform, the difference image (the FPGA acceleration version and normal version) shows a big difference. You can find the the difference image under http://zedboard.org/imce?app=ckeditor%7Csendto%40ckeditor_imceSendTo%7C&CKEditor=edit-comment-body-und-0-value&CKEditorFuncNum=2&langCode=en

Fullscreen contentimage_102087.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

Is there a bug in warpTransform?

0 jalejandro over 7 years ago

I tried attaching the difference image in the original post but it does not show up.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 albertabeef over 7 years ago

The file I/O examples provided with the Avnet UltraZed EG platform, are directly reused from the Xilinx SDSoC/reVISION platforms.
Regarding the warp transform, the implementation is not bit accurate so differences are expected.
How many errors ? this depends on the implementation, and probably the different modes (such as affine versus perspective).
I would recommend taking a look at the Xilinx provided testbench for how many errors should be tolerated.
https://github.com/Xilinx/xfopencv/blob/master/examples/warptransform/xf_warp_transform_tb.cpp
Regards,
Mario.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 cvip over 7 years ago in reply to albertabeef

Hi Mario,

Thank you for your reply.
You can find the differnce image under
http://zedboard.org/imce?app=ckeditor%7Csendto%40ckeditor_imceSendTo%7C&CKEditor=edit-comment-body-und-0-value&CKEditorFuncNum=2&langCode=en

The number of error is 850557.
But it is very interesting. When we use default perspective transform, the number of error is only 477. But when we use different perspective transform, the number of error becomes 850557.

Thanks.

Xianju
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 jalejandro over 6 years ago in reply to cvip

Matrix 1:
Num of error = 477
1.03873 0.00596399 -19.8313
0.00713421 1.0394 -16.7659
1.31407e-05 6.06667e-06 1

Matrix2:
Num of error = 850557
0.936579 -0.0528042 0
0 1.06772 0
-7.21694e-07 -5.98251e-05 1
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel