OpenCV tracking example needs a frame limit

[teh-cmc]

It runs for like two minutes otherwise, making end-to-end tests extremely painful.

Similar to opencv_canny, this should have an optional frame limit.

[javierttgg]

Hi, is there a way to only store the last frame?

Similar to what is achieved with OpenCV by calling cv2.imshow('name', image) multiple times.
This way memory requirements would be reduced since only the most recently processed image needs to be kept in memory.

I think this usage could be very common when visualizing a sequence of images where only the current image is important. However I understand this may not be a use case of the library since we would not be able to rerun the sequence.

I have a similar question for 3D scenes too. Since one of rerun's objectives is #1630, the previous behavior would not only fit imshow and plot but also draw_geometries from Open3D (its viewer only visualizes the current actors with their current transformations).

I feel that this question may fit this issue, but please let me know if that's not the case.

P.S. This is an amazing library, thanks a lot for open-sourcing it!

[Wumpf]

@javierttgg As you inferred correctly, one of the central ideas of Rerun is to store data as you go and don't throw away earlier results. But that said we have a mechanism to reduce the memory used and throw away previous images/logs/etc, see https://www.rerun.io/docs/howto/limit-ram
What we're missing so far there is good defaults, exposition in the api and an easy way to discard "anything older than x".
In my experience the problem isn't as bad for 3d scenes - meshes are relatively lightweight in textures rarely change. Ofc, a large difference is that with Rerun you wouldn't continously call a draw method, but instead re-actively log data if there's something new & interesting.

I feel that this question may fit this issue, but please let me know if that's not the case.

It's very related, but not entirely 😄. The issue is more that we (miss-)use our examples right now a lot for testing and then when running all examples at once and doing so is a bit too lengthy for some, so it would be nice to a have a parameter there to not run through the whole video stream.
But still thank you for posting, any feedback no matter where is very welcome. This confirms that we still need to improve this area quite a bit and to so soon!

This is an amazing library, thanks a lot for open-sourcing it!

Thank you <3

[javierttgg]

@Wumpf thanks a lot for the thorough response. It is great to know that you have those features in mind!
I see now that my question doesn't fit well here, my apologies for adding noise 🙏