There are 3 approaches.
- “tao gazenet inference xxx”. See Gaze Estimation - NVIDIA Docs . For this approach, suggest you to run official released notebook as the starting point. This notebook will download public dataset and run training and inference.
This approach runs in x86 PC only. - Run with deepstream. See https://docs.nvidia.com/tao/tao-toolkit/text/deepstream_tao_integration.html#pre-trained-models-bodyposenet-emotionnet-fpenet-gazenet-gesturenet-heartratenet and https://github.com/NVIDIA-AI-IOT/deepstream_tao_apps/tree/master/apps/tao_others/deepstream-gaze-app
This approach can run in Nano or X86. But currently there is limitation to visualize the gaze vector. DS team is working on that. - Run with an old inference pipeline. This approach is mentioned in previous version of tao. But you can still use it. You can refer to How to visualise the 3d gaze vector output of the GazeNet model? - #27 by Morganh
This approach can work on X86 or Jetson Xaiver or NX.