03-28-2019 01:29 AM
04-01-2019 10:16 PM - edited 04-01-2019 10:24 PM
Yes, they have same functionality, but we have customized openCV functions to work better suited to FPGAs.
The video processing functions included in the HLS Video library are compatible with existing OpenCV functions and are similarly named. They do not directly replace existing OpenCV video library functions. The video processing functions use a data type hls::Mat.This data type allows the functions to be synthesized and implemented as high performance hardware. Refer page age 255 of https://www.xilinx.com/support/documentation/sw_manuals/xilinx2018_2/ug902-vivado-high-level-synthesis.pdf
04-02-2019 01:34 AM
Thank you for your reply.
I have tried to build a demo code where I am calculating mean and standard deviation in HLS scope and openCV also. When comparing it using simulation, I am getting different results. The codes are uploaded at this webpage https://forums.xilinx.com/t5/Vivado-High-Level-Synthesis-HLS/Mean-and-standard-deviation-of-image-using-HLS-and-OpenCV/td-p/953089
When you say that they do not directly replace existing openCV function, does it mean that it will not produce same results?
Also I feel that there is some mismatch in hls::scalar and cv::scalar data type. I have used these 2 datatypes for mean and SD (as per function requirement) respectively for HLS and openCV. Let me know if you could find bug there.
04-02-2019 12:19 PM
Hey @bhavinlapasia ,
If you're using Vivado HLS 2018.3, you should also consider using the xfOpenCV functions. Moving forward, the Vivado HLS video libraries have been moved to the Xilinx GitHub and can be found here: https://github.com/Xilinx/xfopencv