DeepSort and TensorRT

Hello, I’m new with Deepsort and I’m trying to run it on a Jetson Nano

The board has Jetpack 4.4.1 with Pytorch V1.6. I’m able to convert yolo to trt files but I don’t know how to use them withe Deepsort. I tried this repo: GitHub - Stephenfang51/tracklite: Using Tensort to speed up yolov3 with deepsort for MOT but it doesn’t run in my enviroment. Any hint of how should I proceed would be great.

Thanks,

Hi,

Do you mean the DeepSort of the below GitHub?

If yes, the source should be runnable on the Jetson Nano directly.
Have you tried it?

Thanks.

Yes I found a repo that use that one and it actually runs but the FPS is really bad arround 1 FPS. I was wondering if it’s possible to convert tiny-yolo to TRT file then deserealize that file and use it with deepsort for tracking. Currently I can’t find a good tutorial that explain how to merge both things

1 Like

Hi,

Sorry that we don’t have an example like that.
But if the model can export into ONNX, you can inference it with TensorRT.

Thanks.

Hello,

Thanks for your help, I did that and currently I was able to achieve good results.

Hello aldiaz,
I’m trying to do the same thing.
Did you have any problems with pycuda module installation there?

Hi @aldiaz
Can you please share github repo for the same
Thanks in advance

1 Like