Hello
We would like to know whether it’s possible or not to run BodyTrack on Jetson family devices (ARM based) such as Jetson Nano and Xavier NX ?
It might have a significant advantage over BlazePose/Mediapipe that also has 3d pose estimation using only 1 monocular camera.
In addition, it would be really help to have an option to train the models for our custom annotated image DB (similar to this question How can we train and use the custom model for the maxine?).
Thanks
Hi @neilqy60u ,
We don’t currently support Maxine on Jetson devices. That being said, if you are interested in running models on them please check out NVIDIA’s DeepStream SDK.
Link: https://developer.nvidia.com/deepstream-sdk
Link: NVIDIA DeepStream SDK Developer Guide — DeepStream 6.1.1 Release documentation
Hi @tvarshney
As far as I know and according to Nvidia Documentation (DeepStream and TRT_Pose) these are 2D human pose estimators but not 3D like Maxine AR BodyTrack. Am I mistaken somewhere ? We are interested in 3D joint positions but not 2D that is why I am asking about Maxine support on Jetson boards.
Another question. Can TRT_Pose/DeepStream use Maxine AR BodyTrack models (.trtpkg) to estimate 3D human pose? TRT_Pose uses .pth files and according to its documentation
“Training scripts to train on any keypoint task data in MSCOCO format. This means you can experiment with training trt_pose for keypoint detection tasks other than human pose.”
Thanks
https://developer.nvidia.com/blog/creating-a-human-pose-estiation-application-with-deepstream-sdk/
Might be I found an alternative here: GitHub - RArbore/Deep-Learning-Hearing-Aid: Various components for using an FCNN to de-noise audio in real-time on the Jetson Nano 2GB. . Anyway it has an issue, the model “model.pt” is lacking in the repository.