When I use deepstream for vision object detection, the lidar detection model without deepstream get wrong

In Jetson orin machine, my deepstream version is 6.1. When I use deepstream for camera object process, the lidar detection model used by tensort send out the error result. I don’t know how to solve the problem, Do any one have some idea? Thank you very much

What kind of lidar detection model? What is the input of the model? Why do you mention you use camera object process with the lidar model?

We already have Lidar data(pointpillar) inferencing with pointpillar inferencing model PointPillarNet | NVIDIA NGC since DeepStream 6.2 GA. The input of the model is pointpillar data but not camera video or images.

The kind of lidar detection model is lidar object detection model. The input for detection model is pointpillar. I use camera object process with camera object model. But at the same time, I run lidar object detection model by tensorrt. The camera object model run by deepstream work correctly, but the lidar object detection model run by tensorrt ran wrongly.

We don’t know anything about your lidar object detection model. So we can not give any comments on your problem.

There is Nvidia TAO lidar model DeepStream sample in the DeepStream SDK 6.2 GA version, please find the sample code in /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-lidar-inference-app, it works with the Nvidia TAO lidar inferencing model PointPillarNet | NVIDIA NGC

The related document: DeepStream Lidar Inference App (Alpha) — DeepStream 6.2 Release documentation

The deepstream framework might have some conflicts with ordinary tensorrt framework. When I initialize ordinary tensorrt model first, then I initialize deepstream pipeline. The deepstream pipeline run correctly, but the tensorrt model sometimes run wrongly(sometimes run correctly). It seems the problem occur in the initialization period.
How can I fix the problem?

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

What do you mean by “initialize ordinary tensorrt model”? We can not get any useful information just from this simple description. Can you provide a simple sample which we can run to reproduce the problem?

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