How to Use “model-instance-id" about dlstreamer HOT 4 CLOSED

I was searching for "model-instance-id" in this repo to see more examples and documentation (https://github.com/search?q=repo%3Adlstreamer%2Fdlstreamer%20model-instance-id&type=code).

Can you try with the simpliest possible pipeline like using e.g. "samples/gstreamer/benchmark/benchmark_two_models.sh", starting with one of the given sample models (using simple default model-proc files)?

Are you sure your used video-files (yes, both are decoded via vaapih264dec) result in the same format?
Because with the above search I saw an error message text saying

All image formats and memoryType for the same model-instance-id in multichannel mode must be the same.

from dlstreamer.

hi @brmarkus, thanks for your advice.

I have tried using the '&' symbol to connect multiple pipelines, where each pipeline has the same model with model-instance-id set to the same value. As mentioned earlier, "When we set the same model-instance-id for the model in multiple gst-launch processes and run these pipelines in parallel, we observe that the GPU memory does not decrease as expected."

However, the GPU memory usage is similar to when model-instance-id is not set. Therefore, I feel that it is not having the intended effect.

from dlstreamer.

Do you see both scripts, benchmark_two_models.sh and benchmark_one_model.sh, running - starting with the default values, default models, default video files, to make sure the pipelines are working, parameters and videos are "compatible", for consistency-check?
Do you see a correlation with the used memory consumption between running "one model" and "two models"?
Would you have a chance to compare with running on the CPU?

from dlstreamer.

Closed due to age an no further response from submitter

from dlstreamer.