patterson163/TensorRT-PeopleNet

Change NVidia TLT(Transfer Learning Toolkit) PeopleNet To TensorRT and Run TensorRT Inference

How to Change TLT PeopleNet to TensorRT type & Run on Jetson(Nano) in Python

Documents

PeopleNet NVidia Address

https://ngc.nvidia.com/catalog/models/nvidia:tlt_peoplenet

PeopleNet TLT Model Download Address

https://ngc.nvidia.com/catalog/models/nvidia:tlt_peoplenet/files?version=pruned_v2.0

Environments & Tools

• A Jetson Hardware(Nano,TX2,TX1,Jetson AGX Xavier,Jetson Xavier NX)

• Jetpack(On Jetson Hardware,you should have installed a jetpack which version is V4.3 or V4.4 or V4.4.1)

• tlt-converter(Depending on different versions of Jetpack)

Steps

1. Download

• Download PeopleNet model

PeopleNet Model Address

• Download tlt-converter in Jetson platform

tlt-converter in Jetson Platform Address

If you run tlt-converter in Jetson platform, we support three kinds of tlt-converter.
https://developer.nvidia.com/tlt-converter-trt51
https://developer.nvidia.com/tlt-converter-trt60 4
https://developer.nvidia.com/tlt-converter-trt71 4

2. Use tlt-converter change TLT-type(.etlt) to TensorRT-type

./tlt-converter resnet34_peoplenet_pruned.etlt -k tlt_encode -d 3,544,960 -t fp1

3. Use TensorRT to Inference

cd sources
./main.py

After you execute the main.py, you will get a output image as below.

Input Image & Output Image

Input Image

Output Image

If this is useful for you,please star me.