Issue with YOLOv11 Models in Deepstream Python Pipeline - Bounding Boxes Only Appear for Primary GIE

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Details of Setup

  • Hardware Platform (Jetson / GPU): GPU
  • DeepStream Version: 7.1.0
  • TensorRT Version: 10.7.0.23-1+cuda12.6 amd64
  • NVIDIA GPU Driver Version (valid for GPU only): 560.35.03 (CUDA Version: 12.6)

Environment:

  • DeepStream 7.1
  • YOLOv11 custom parser compiled for NVIDIA TensorRT
  • Ubuntu 22.04 LTS, Dockerized deployment

Issue Type: Bug

Description of Issue:

I am currently working on a DeepStream python pipeline that utilizes two YOLOv11 models. The setup includes one primary GIE for car detection and another secondary GIE model for additional object detection tasks.

When each model is run individually as the primary GIE, they work correctly and display bounding boxes as expected. However, when both are integrated into a pipeline with one acting as the primary GIE and the other as the secondary GIE, only the bounding boxes from the primary model appear in the output. The secondary model’s detections do not show up despite being configured similarly.

I have reviewed the following discussion for similar issues:

Based on the suggestions in these threads, I have ensured that my configuration files are correctly set up and that there are no apparent syntax errors or misconfigurations. However, the problem persists.

Any insights or advice on how to resolve this issue would be greatly appreciated.

For more information see my earlier solved topic about my setup: Issue with YOLOv11 Models in DeepStream 7.1 - #4 by mazewalker

3

So what is your problem now? You can’t detect the license plate? In your previous topic, wasn’t it already possible to detect it?

Is it just the difference between your python program and deepstream-app?

4

License plate gets detected when set as Primary GIE and bounding box shows up. Same with car detections (works as Primary GIE). When car detection is in Primary GIE and license plate detection is in Secondary GIE then only cars are being detected, but license plates aren’t.

5

Please refer to this sample deepstream_reference_apps/legacy_apps/back-to-back-detectors at master · NVIDIA-AI-IOT/deepstream_reference_apps · GitHub

Particular, for primary detector

process-mode=1

For secondary detector

process-mode=2
# The id of the vehicle class
operate-on-class-ids=2
1 Like
6

The issue was not in the configs, but quite simply a misplaced recently converted ONNX file on the development server caused the issue.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.