secondary inference parallelism

Let’s say I have a common deepstream pipeline where the inference part is made of a detector (pgie) and three classifiers (sgie1, sgie2, sgie3), connected in this fashion.

gst_element_link_many(<some source>, streammux, pgie, sgie1, sgie2, sgie3, <some sink>, NULL)

I would like to understand if there is any parallelism between the secondary inference operations, since they appear to be working sequentially, but their input in fact depends only on the output of the pgie.
Can sgie1 access data and metadata at the same time as sgie2, and sgie3? If so, how is this achieved?

Thanks in advance,


Hi, you needs to implement the app which adds a tee after the PGIE and link the 3 SGIEs to the tee. You can refer to DeepStream App for this.