NvDCF Tracker is not idempotent

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi team,

We are currently using DeepStream 5.1 and performing person detection with line crossing. If we run it on large number of videos with high density of persons, the counts of subsequent runs do not match. There are 5 - 10% difference in counts every time we run on same set of videos

I am wondering if this is a known issue, and how can we avoid it?

• Hardware Platform: Jetson NX and T4 GPU
• DeepStream Version: 5.1
• JetPack Version: 4.5.1
• TensorRT Version: 7.2.1

3

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

Can you provide the steps and video to reproduce this problem?