Hi all, I’m currently trying to make an OCR pipeline and have some problem with post-processing speed.
So, I want to replace all cpu methods in post-processing to gpu methods.
My post-process pipeline needs to predict connectedComponent since the output of neural-network is 1d-segmentation image.
Yeah, as you know, there’re a solution for CCL(Connected Component Labelling) like
npp-label-marker
YACCLAB
But I need to use a method to find min-area-rect (like cv2:minAreaRect()) after CCL.
I found some slides about this problem from GTC2019 but cannot find any source codes about CCA(Connected Components Analysis)(https://developer.download.nvidia.com/video/gputechconf/gtc/2019/presentation/s9111-a-new-direct-connected-component-labeling-and-analysis-algorithm-for-gpus.pdf)
Is there any code which can use CCA directly?
Or should I make some codes for it?
If so, can I make it with pycuda?
Any ideas…?
Thank you all. Sincerely yours.