• Jetson AGX Orin • DeepStream 7.0 • JetPack Version 6.0 • TensorRT Version 8.5.2.2 • Question
Hi! I’m trying to use this deepstream-3d-pose as my Secondary GIE (SGIE). I’ve added the config file for the SGIE, but it doesn’t show any keypoints in the output. Since there’s no Python binding example for this pose estimation model, I wonder if it’s possible to use it with Python binding? Why doesn’t my app output the keypoints? What process might I be missing? Are there any references or guidance available for using the pose estimation model with Python binding? Any help or direction would be greatly appreciated!
first please make sure pgie can output the bboxes of peoples. you can reuse the config_infer_secondary_bodypose3dnet.txt. then you get the sgie’s inference results in sgie’s probe function. you need to process inference data and add display meta. please refer to code.
I see, so I need to ensure my primary GIE includes a person class, use the config_infer_secondary_bodypose3dnet.txt as the secondary config file, and add the sgie_src_pad_buffer_probe to handle and display the outputted keypoints from the pose estimation. Is that correct?
Right now I’m trying to create the sgie_src_pad_buffer_probe function with python binding, I refer to the cpp file as you suggested. During the process, I noticed there’s a _pose_file in the sgie_src_pad_buffer_probe. What is that? Will the output of the pose estimation be a file?
about “Is that correct?”, yes. you need to debug step by step.
about "What is that? " , no, _pose_file is used to dump the points to a file. you can ignore it if you don’t need the file. inference results are in “NvDsInferTensorMeta *tensor_meta = (NvDsInferTensorMeta *)user_meta->user_meta_data;”, please refer to parse_25dpose_from_tensor_meta for how to process.
There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks