No.
Deepstream is designed for common use for all kinds of videos inferencing, and no one can predict how many objects will be detected by primary detector. The batch size of SGIE is a tradeoff. And the SGIEs also using GPU resources, it can not reach the performance of single detector in the pipeline.