Batch-Size confusion


I’m a bit confused about the different batch sizes.
There is one for nvstreammux, one for each instance of nvinfer, one for each model config file, one for tracker (ReId model).

In a multi-stream environment, with detectors and several classifiers, I’m having trouble understanding how to configure these batch sizes.

Could you define each of them ?

I don’t want to build an engine every time I have 2;3;4;…;10 streams. Can I build a batch-size=100 engine and use it for 5 streams? Does this make sense and will it be optimal?

Yes. You can do this. But it is more efficient to set batch-size to the number of your sources.

Can you explain why?
Can you define all other batch-size mentioned in initial post?

This allows a batch to process each frame of data from each source.

You can refer to our Guide about that.