Hello guys,
I have a usecase to add and remove running models on a fly. Can it be done with a Holoscan?
Imagine that I have a video input, I want to apply 2 models in parallel to the frames. I build a graph, make an app and run it asynchronously.
And then during runtime another one model should be added: now I need to run 3 models in parallel. And I do not want to lose frames of course. So, my questions are:
- is it possible to add a new model to the graph or to the Inference operator without stopping the running app?
- if it is not possible, then if I stop app, rebuild the graph and run app once more, how much time may it take on Jetson AGX Xavier (how many frames will I lose).
I can imagine a solution: to build a new graph with 3 models, run the corresponding app asynchronously in parallel to the existing app. Once the new app is up and running, I remove the first app. But that looks a little bit srange in terms of pike resources consumption.
With the best wishes,
Valeriy