Hand Posenet From NVIDIA-AI-IOT / trt_pose_hand Demo

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Is there a way to get the a trainable Tao model of the hand pose model (
hand_pose_resnet18_att_244_244.pth) used in GitHub - NVIDIA-AI-IOT/trt_pose_hand: Real-time hand pose estimation and gesture classification using TensorRT ?
I would like to do some transfer learning.

3

Actually it is similar to gesturenet in TAO. Please download notebook from

or GPU-optimized AI, Machine Learning, & HPC Software | NVIDIA NGC

The pretrained model is mentioned in https://docs.nvidia.com/tao/tao-toolkit/text/purpose_built_models/gesturenet.html
→ GestureNet | NVIDIA NGC
https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/models/gesturenet/files?version=trainable_v1.0

Then run gesture training in TAO.

1 Like
4

gesturenet seems to be a classification model. Is it possible to just get the 2d hand pose model?

5

Sorry, currently TAO does not have a hand pose model. It has body pose model instead. https://docs.nvidia.com/tao/tao-toolkit/text/purpose_built_models/bodyposenet.html
I will sync with internal team for it.

1 Like
6

If I use the body pose model, is there any way to get the exact data and parameters that were used when training hand_pose_resnet18_att_244_244.pth internally?

7

The hand_pose_resnet18_att_244_244.pth is not compatible with TAO.
For TAO bodypose model, please refer to its trainable model BodyPoseNet | NVIDIA NGC
End users can use it to train.
Or use the deployable model directly to run inference.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.