Video Processing API - processing inside docker on Jetson is noticeably slower than bare-metal

Search before asking

• I have searched the Inference issues and found no similar bug report.

Bug

The same workflow tested, reporting only latency for single frame processing inside WorkflowRunner.run_workflow(...) function:

Jetson Nano Orin, bare metal in script using InferencePipeline directly - not measured precisely now, but older tests indicated the same performance as on MacBook when yolov8n-640 used - which was the model used in test case
Jetson Nano Orin, inside docker container, behind API - ~50ms
We have docker overhead, not 100% sure if it is visible on Jetson devices, but MacBook one makes drop from 27fps into <10fps 😢

Also measured that running 2 streams at the time gives over 30FPS in total so the compute is there, but something makes the pipeline slower than potentially could be inside Jetson Docker

Environment

Jetson Nano Orin, Jetpack 5.1.1

Minimal Reproducible Example

No response

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

Start with inference/core/interfaces/stream/model_handlers/workflows.py and the WorkflowRunner.run_workflow(...) path, then reproduce the single-frame latency comparison on Jetson Nano Orin with JetPack 5.1.1 in bare-metal and Docker environments. Done means the container slowdown is explained and the processing performance is improved or the relevant limitation is documented with measured results.