Second inference interval not working (original) (raw)

April 28, 2025, 11:36am 1

• Hardware Platform (Jetson / GPU)
GPU
• DeepStream Version
7.1 Python SDK
• TensorRT Version
tensorrt-dev 10.3.0.26-1+cuda12.5
• NVIDIA GPU Driver Version (valid for GPU only)
570.86.10
• Issue Type( questions, new requirements, bugs)
Question

I am running a simple pipeline using Deepstreamd Python SDK 7.1 with 2 models nvinfer inference, a person detector and a reid model as sgie. I also have a probe after the second inference where i read the bbox, image and reid embeddings.

uridecodebin → nvstreammux → queue → nvinfer1-> queue → nvtracker → queue → nvinfer2 (SGIE) → queue → nvvideoconvert → queue → capsfilter → queue6(+ probe) → fakesink
The probe is attached to the sink pad of queue6 before fakesink.

I want to only perform inference on some frames, so we I set the interval to skip inference in the first model for 5 frames and on the second model we skip 30 frames. I setup the config parameter interval as seen in the config files.

My problem is that in the probe I keep reading for every frame the bbox and reid embeddings data. I tried checking if the data was just repeated from a prev. frame if we skip that frame but every frame I get different embeddings.

Does that mean the interval setting is not working?

Could you please check my settings and let me know if I should do something different?

detector_config.txt (414 Bytes)
tracker_config.txt (1.2 KB)
reid_config.yml.txt (530 Bytes)

fanzh April 29, 2025, 1:52am 3

please use secondary_reinfer_interval for sgie instead of interval. please find the explanation in the doc.