Parsing output of PeopleSegNet with Tensorrt in python

Accelerated Computing Intelligent Video Analytics TAO Toolkit
,
1

Hi,

I am trying to run inference on PeopleSegNet tensorrt engine file that i was able to get by using tlt-converter tool on the pretrained .etlt model i got from here
on Jetson xavier AGX.

I had no problem with the conversion, now i am trying to run the inference on the exported engine file in python.
But i can’t seem to make sense of the output, i believe i might be doing the pre-processing or post-processing wrong.

I am not sure what i am doing wrong though.

This is the script i use for inference in python
PeopleSegNet_trt.py (1.1 KB)

Any help would be appreciated

P.S that i ran trtexec on the engine file and it ran fine, i also tried it with deepstream and it worked fine as well, which is why i believe the problem is in the pre-processing or post-processing.

3

In addition, I read here about the output shapes but i still don’t know how those outputs can be converted or processed in any way to be usable directly with opencv to draw the bounding boxes and masks.

4

For preprocessing, it will keep aspect ratio and then resize and pad. Reference:Inferring Yolo_v3.trt model in python - #26 by Morganh

For how to process the trt engine, when you read GitHub - NVIDIA-AI-IOT/deepstream_tao_apps: Sample apps to demonstrate how to deploy models trained with TAO on DeepStream, you can also refer to other user’s experience in Inferring Yolo_v3.trt model in python (for yolo) and Inferring detectnet_v2 .trt model in python - #46 by Morganh (for yolo and detectnet_v2) and Run PeopleNet with tensorrt - #21 by carlos.alvarez (for detectnet_v2)

5

Thanks alot