Skip to content

re-implementation of diffusion trajectory generation for edge devices

Notifications You must be signed in to change notification settings

saucesaft/nomadrt

Repository files navigation

nomadrt (real time)

implementation of NoMAD with tensorrt for running on edge devices. tested to be around 6x faster than the pytorch code.

convert from pytorch to onnx

python onnx_export.py -i nomad.pth -t config/nomad.yaml -o weights

convert onnx models to tensorrt's engine

cd weights
trtexec --onnx=action.onnx --best --useCudaGraph --saveEngine=action.engine
trtexec --onnx=encoder.onnx --best --useCudaGraph --saveEngine=encoder.engine
trtexec --onnx=distance.onnx --best --useCudaGraph --saveEngine=distance.engine

compile the ros2 package

colcon build --symlink-install

launch the demo node

ros2 launch nomadrt video_demo.launch.py

special thanks

About

re-implementation of diffusion trajectory generation for edge devices

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published