Skip to content
Discussion options

You must be logged in to vote

I agree with the ideas from @alexriedel1
Additionally, it is better to go with optimised exported models like onnx with onnx runtime instead of pure torch.
OpenVINO (although primarly used for x86 , also supports ) is another option but better to benchmark it on the target device.

4K frame processing sounds a lot especially for on 2GB memory.
On the image size :

  • There could be a high chance that you do not need to process images at 4K high resolution. You could test with lower image resolutions and find the sweet spot where you do not loose any accuracy.
  • Depending on the problem, you might also do cropping to a fixed region of interest before inference

About the tiling part,
You could a…

Replies: 1 comment 2 replies

Comment options

You must be logged in to vote
2 replies
@rajeshgangireddy
Comment options

Answer selected by acode-x
@acode-x
Comment options

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
3 participants