I found that your sample 04 video_dec_trt deep copy the dec_capture_fd to dst_dma_fd(egl_imagePtr), why you do it like this? Can I map the dec_capture_fd to egl_imagePtr (zero copy) then transform and the infer it by cuda? Or the method dec_capture_fd to egl_imagePtr will make the performance lower? Or some errors?
Hi,
Decoded YUVs are in block linear so we will convert it to pitch linear for later processing. This is done on VIC engine and please run the script to get maximum performance:
VPI - Vision Programming Interface: Performance Benchmark
That means I can not NvAllocate the decCaptureFd to NVBUF_LAYOUT_PITCH
I must NvAllocate the decCaptureFd to block linear and transform it to NVBUF_LAYOUT_PITCH, right?
Hi,
This is correct.
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.