Please provide complete information as applicable to your setup.
**• AWS EC2 g4dn.xlarge
**• DeepStream 5.1
**• TensorRT 7.2.1.1
**• NVIDIA-SMI 455.23.05 Driver Version: 455.23.05 CUDA Version: 11.1
On the Deepstream 5.1 Release notes document, there is point regarding NvDCF tracker enhancements in DS 5.1. What are those enhancements?
Anything related to the tracker yml config file?
Thanks and regards
Hey, we will check internally and reply you ASAP.
Great, thanks!
Hello, the enhancements in DS 5.1 for NvDCF tracker is mostly on performance and robustness.
Anything related to the tracker yml config file?
There’s no major change in config file structures or so. Do you have particular question or request about the config file?
Thanks for your feedback. We ask because we had an application built on DS 5.0 and the same application with the same config file on DS 5.1 has a worse performance. Could you please be more specific and let us know what exactly has been improved (in general) in DS 5.1?
Thanks again and regards
Hey, I had compared the samples/configs/deepstream-app/tracker_config.yml from DS5.0.1 and DS5.1, there are some minor change, could you try to use the tracker_config.yml in DS5.1 and see if performance better
diff /home/bcao/customer-bug/tracker_config.yml.5.0.1 /home/bcao/customer-bug/tracker_config.yml.5.1
58a59
> usePersistentThreads: 0 # If true, create data association threads once and re-use them
64,65c65
< minMatchingScore4VisualSimilarity: 0.2 # Min visual similarity score
< minTrackingConfidenceDuringInactive: 1.0 # Min tracking confidence during INACTIVE period. If tracking confidence is higher than this, then tracker will still output results until next detection
---
> minMatchingScore4VisualSimilarity: 0.2 # Min visual similarity score
We have already tried the DS 5.1 tracker_config.yml but the performance is not better. Thanks anyway, we will keep on investigating