Please provide complete information as applicable to your setup.
**• Hardware Platform -------> GPU
**• DeepStream Version -------> 6.4
**• TensorRT Version -------------> 8.5
**• NVIDIA GPU Driver Version ----------> 545
I am experimenting with “sub-batches” tracker feature,
I have 10 stream and each stream having 100 object,
what I’ve done is each stream I created one sub-batches (10 stream , 10 sub-batches). while running after some time I’m getting this error
“gstnvtracker: ConvBuf status is not available for active batch!
gstnvtracker: completed req is not in active batch!
Error: gst-stream-error-quark: Failed to submit input to tracker (1), gstnvtracker.cpp(706): gst_nv_tracker_submit_input_buffer ():”
Why this error is coming.
I want to use sub-batches very wisely where performance and accuracy will be balance.
When I have set sub-batches 3 ( 3streams, 3streams, 4 streams). It’s working good.
How do I measure that sub-batches per-stream will work good.
If you have list or tested measurement please share that !
I’m using 6.4 as of now, Why you are saying to upgrade ? is batch process is not available in 6.4 ?
I did not see any different documentation change between 6.4 vs 7.0 for sub-batches and tacker.
I’m attaching my pipeline graph as well.
Yes, it’s available in 6.4 version. But if you want to use deepstream-app, sub-batch parsing was not added in the deepstream-app source code in DS 6.4.
Could you remove the bracket for sub-batch parameter in the config file?
In Deepstream documentation
“”“”
Option 1 : Semicolon delimited integer array where each number corresponds to source id.
Must include all values from 0 to (batch-size -1) where batch-size is configured in [streammux].
Option 2 : Colon delimited integer array where each number corresponds to size of a sub-batch (i.e. max number of stream a sub-batch can accommodate)
“”"
tracker.txt
here it’s not taking the config, I’m setting the property like this
new_model_element.set_property(‘sub-batches’,“0;1;2;3;4;5;6;7;8;9”)
It’s running But after some time it’s showing above error and stopping. I want understand sub-batching with proper utilisation, and one sub-batch how many stream can handle and what will be the appropriate way to use batch-processing.
Can u help us to understand how many sub-batches can be supported by one tracker,
Pls provide us performance metric so that we can understand should we go with sub batches or multiple tracker instances fro one sgie.
As more than 4 sub-batches in reltime application is giving the error
gstnvtracker: ConvBuf status is not available for active batch! gstnvtracker: ConvBuf status is not available for active batch!
gstnvtracker: completed req is not in active batch!
gstnvtracker: completed req is not in active batch!
We have refer to this code “gst-plugins/gst-nvtracker/nvtracker_proc.cpp”
Pls look into this
We added an illustration of the sub-batching examples in the nvtracker_proc.cpp on DS 7.0. You can refer to that.
We don’t have a performance metric at the moment. This parameter should be configured according to the GPU loading of your device.
Could you attach your video source with 100 objects and the specifies your GPU model? We can try that on our side.
You can also monitor your GPU loading and adapt the appropriate value based on your GPU loading.
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
