If you are using the model from Pose Classification | NVIDIA NGC, the output " fc_pred" is just logits values, you need to do softmax algorithm to get the real probabilities.
Hi @Fiona.Chen ,
In order to implement Softmax I am trying to understand the output of the variables in “NvDsInferClassifierParseCustomSoftmax”.
I have added standard print statement in the code and again created .so file .
std::cout<<“outputCoverageBuffer”<< outputCoverageBuffer << std::endl;
std::cout<<“Probability”<< probability << std::endl;
When I try to run the pipeline again, I can’t see any output regarding outputCoverageBuffer and Probability being displayed on the command terminal.
Could you please let me know how should we the print the values of the variables in the NvDsInferClassifierParseCustomSoftmax function.
Thanks
It looks like you need to make sure that preprocess and sgie2 are added to the pipeline correctly and that the play state is correct.
Can you connect a buffer probe to sgie2 and not tiler to check the standard output?
Have you make sure you have generate correct input to the model? How did you generate the input tensor data?
We are developing the sample for pose classifier model. Please wait for the next DeepStream release.
We are developing the sample for pose classifier model. Please wait for the next DeepStream release
Hi @Fiona.Chen , Sorry for the late reply. I was not available. Could you please let me know when is the next DeepStream Release expected?
Have you make sure you have generate correct input to the model? How did you generate the input tensor data?
No, what we understood is that After implementing Sgie1 ( BodyPose3d Net) we will get the output as 34 * 4 and before implementing SGIE2 ( PoseClassificationNet), we need to create a preprocess function to convert 34 * 4 output into (N, C, T, V, M) format . I am not proficient in C nor I found any reference which I can refer to create the preprocess function.
If there is any reference code / script , Could you please share with me?
Thanks,
Shabbir
We will announce in the forum once the new version is available.
It is not a simple piece of codes. Please wait for the release.
Deepstream is released, details on link DeepStream 6.3 is now available for download! - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.