How to setup trt_pose for Jetson Orin Nx.
- Check Jetpack version.
sudo apt-cache show nvidia-jetpack
- Install corresponding pytorch following this link
sudo apt-get install python3-pip libopenblas-dev
pip install <wheel-name>
or after installing python run:
pip3 install torch torchvision
- After installing torch and torchvision, install torch_2_trt
git clone git@github.com:NVIDIA-AI-IOT/torch2trt.git
pip3 install packaging
<set CUDA_HOME var to the location of your cuda installation [export CUDA_HOME=/usr/lib/aarch64-linux-gnu/]>
python3 setup.py install --plugins
** NOT SUCCESSFUL --> FOR NOW **
Download jetson-containers.
Run exemplary jetson container
jetson-containers run $(autotag l4t-pytorch)
It doesn't assign container name (that's wanted) so it is needed to check name of the container:
enter container and then install torch2trt
from github repo (as stated above).
After running jetson-containers (remove rm
flag in the command),
build trt_pose
and build torch2trt
.
First install torch2trt
as follows:
git clone https://github.com/NVIDIA-AI-IOT/torch2trt.git
cd torch2trt
python3 setup.py install
python3 setup.py install --plugins
After installing torch2trt install trt_pose
:
git clone https://github.com/NVIDIA-AI-IOT/trt_pose.git
python3 setup.py install
Install notebook
and tqdm
pip install notebook tqdm pycocotools
Follow README.md
from trt_pose
to see where to put weights of an NN.
Problem with import trt_pose.plugins
can be solved using following link.
- Explore jetson containers
- Run contaier from
jetson-containers
- Check
trt_pose
demo in ROS node - Add
trt_pose
to ROS node - Test ROS 2 humble container
- Add
trt_hand_pose
- Create Dockerfile that's used to setup correct env for HPE on Jetson --> Not possible
- Create script that set ups fast from jetson-containers base image
setup.sh