Multicamera detection issues in deepstream 7.0 with jetson

When running DeepStream 7.0 for multi-camera setups, detection is only performed on some of the cameras.

The same app works fine for multiple cameras on a cloud machine using deepstream 6.1 on Ubuntu 20.04. However, it was necessary to update deepstream to 7.0 to use Jetson on Ubuntu 22.04.

The application is running on a Jetson device with 16GB RAM. With one camera, detection works perfectly. With two cameras, detection starts failing partially on one of them. With four cameras, detection occurs on only some cameras, even under identical conditions.

There are two models: one for person detection and another for safety equipment detection.
The model weights and configuration files are listed below:

equipment model configuration file (699 Bytes)
people model configuration file (813 Bytes)

equipment model (45.6 MB)
people model (46.0 MB)

Models were converted to .onnx using the Ultralytics library and then converted to .engine with the configuration file below:

yolov8_ubuntu22_jetson.zip (276.4 KB)

The templates were called in the following yaml configuration file.

app_config.zip (1.7 KB)

The same application works well for multiple cameras on a cloud machine using deepstream 6.1.

• Hardware Platform (Jetson / GPU) : NVIDIA Jetson Orin NX Engineering
• DeepStream Version : 7.0
• JetPack Version (valid for Jetson only) : 6.0
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only) : 540.3.0
• Issue Type( questions, new requirements, bugs) : Bugs
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)

Could you dump and attach one of your short video source? Also can you simplify your project so that we can try to reproduce it on our devices?

The project consists of two models, one for detecting people and the other for detecting the absence of safety equipment, the labels for the second model are below. Models are run independently of the image and people and equipment are matched after detection. Risk events are generated when the absence of equipment is detected in areas of interest.

lequipment model labels (80 Bytes)
short video (9.6 MB)

I am also sending the converted models to onnx
Onnx person model (82.5 MB)
Onnx equipment model (81.9 MB)

Could you try to use FP32 mode by setting the network-mode to 0?