3D human pose estimation - OpenVINO Error: Demo supports topologies only with 2 outputs

[thunfischtoast]

Hi!

I have the same problem as https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/OpenVINO-Error-Demo-supports-topologies-only-with-2-outputs/m-p/1292253 however the question was not really answered there.

I have setup the https://docs.openvinotoolkit.org/latest/omz_demos_human_pose_estimation_demo_cpp.html as instructed, however when I try to run the demo on a test video I get the error

C:\Users\croemer\Documents\Intel\OpenVINO\omz_demos_build\intel64\Release\human_pose_estimation_demo.exe -m "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\public\human-pose-estimation-3d-0001\FP16\human-pose-estimation-3d-0001.xml" -i C:\Users\croemer\data\raw\videos\v3\1595278991805.mp4 -at openpose
[ INFO ] InferenceEngine:       IE version ......... 2021.4
        Build ........... 1
[ INFO ] Parsing input parameters
[ INFO ] Reading input
MFX: Unsupported extension: C:\Users\croemer\exchange\fw_webapp_backend\data\raw\videos\v3\1595278991805.mp4
[ INFO ] Loading Inference Engine
[ INFO ] Device info:
[ INFO ]        CPU
        MKLDNNPlugin version ......... 2021.4
        Build ........... 1
Loading network files
[ INFO ] Batch size is forced to 1.
[ ERROR ] Demo supports topologies only with 2 outputs

The same error occurs when I just define a single jpg as input.

Defining 'ae' as network type actually yields a different error:

C:\Users\croemer\Documents\Intel\OpenVINO\omz_demos_build\intel64\Release\human_pose_estimation_demo.exe -m "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\public\human-pose-estimation-3d-0001\FP16\human-pose-estimation-3d-0001.xml" -i C:\Users\croemer\data\raw\videos\v3\1595278991805.mp4 -d GPU -at ae
[ INFO ] InferenceEngine:       IE version ......... 2021.4
        Build ........... 1
[ INFO ] Parsing input parameters
[ INFO ] Reading input
MFX: Unsupported extension: C:\Users\croemer\exchange\fw_webapp_backend\data\raw\videos\v3\1595278991805.mp4
[ INFO ] Loading Inference Engine
[ INFO ] Device info:
[ INFO ]        GPU
        clDNNPlugin version ......... 2021.4
        Build ........... 1
Loading network files
[ INFO ] Batch size is forced to 1.
[ ERROR ] output layers are expected to have 1 batch size and 17 channels

and the same with 'higherhrnet'

Can you please give me advice on steps to narrow down the problem?

Thank you!

[Wan_Intel]

Hi Thunfischtoast,

Thank you for reaching out to us and thank you for using the OpenVINO™ toolkit!

I noticed that you are using human-pose-estimation-3d-0001 for Human Pose Estimation C++ Demo. For your information, human-pose-estimation-3d-0001 is not a supported model for Human Pose Estimation C++ Demo. Because of this, you will encounter the error: Demo supports topologies only with 2 outputs.

Supported models for Human Pose Estimation C++ demo are as follow:
 

·      architecture_type = openpose

o human-pose-estimation-0001
 

·      architecture_type = ae

o human-pose-estimation-0005

o human-pose-estimation-0006

o human-pose-estimation-0007
 

·      architecture_type = higherhrnet

o higher-hrnet-w32-human-pose-estimation

For example, to run Human Pose Estimation C++ Demo and do inference on a CPU, please run with the following command:

./human_pose_estimation_demo \
-i <path_to_video>/input_video.mp4 \
-m <path_to_model>/human-pose-estimation-0001.xml \
-d CPU \
-at openpose

On another note, human-pose-estimation-3d-0001 is a supported model for 3D Human Pose Estimation Python Demo. To run 3D Human Pose Estimation Python Demo, please run with the following command:

python human_pose_estimation_3d_demo.py \
-m <path_to_model>/human-pose-estimation-3d-0001.xml \
-i <path_to_video>/video_name.mp4

Hope it helps.

Regards,

Wan

[thunfischtoast]

Hi Wan,

thank you very much for the clarification and help!

I actually missed the part where I lost the "3d" on the .exe path

However unfortunately the python model doesn't load for me:

I built the demos as instructed using 

C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\build_demos_msvc.bat -DENABLE_PYTHON=YES

and the build ran without errors.

The output is saved at C:\Users\<my_user>\Documents\Intel\OpenVINO\omz_demos_build\intel64\Release

However when I try to run

python3 "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\human_pose_estimation_3d_demo.py" -m "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\public\human-pose-estimation-3d-0001\FP16\human-pose-estimation-3d-0001.xml" -i C:\Users\<my_user>\data\raw\videos\v3\1595278991805.mp4 -d GPU

 (the multi-line command with \ at the end actually does not work for me in cmd)

I get

Traceback (most recent call last):
  File "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\human_pose_estimation_3d_demo.py", line 25, in <module>
    from modules.parse_poses import parse_poses
  File "C:\Program Files (x86)\Intel\openvino_2021\inference_engine\demos\human_pose_estimation_3d_demo\python\modules\parse_poses.py", line 20, in <module>
    raise ImportError("Module 'pose_extractor' not found. "
ImportError: Module 'pose_extractor' not found. Please build module according to documentation before usage.

I've built a python venv with the openvino-dev version 2021.4.1 and ran setupvars.bat to set up the environment and added C:\Users\<my_user>\Documents\Intel\OpenVINO\omz_demos_build\intel64\Release to the PYTHONPATH using 

set PYTHONPATH="C:\Users\<my_user>\Documents\Intel\OpenVINO\omz_demos_build\intel64\Release\;%PYTHONPATH%"

The Release-folder contains pose_extractor.exp, .lib and .pyd

Can you give me a pointer where the pose_extractor module is supposed to be?

Thank you!

Cheers

[Wan_Intel · Accepted solution]

Hi Thunfischtoast,

Thanks for your information.

Great! You have successfully built the native Python extension module for 3D Human Pose Estimation Python Demo. As you mentioned, pose_extractor.exp, pose_extractor.lib, and pose_extractor.pyd were located under the Release folder.

For your information, pose_extractor module is located under the following directory:

<INSTALL_DIR>\deployment_tools\open_model_zoo\demos\human_pose_estimation_3d_demo\python

Before you run the demo, you must copy pose_extractor.exp, pose_extractor.lib, and pose_extractor.pyd into the following directory:

<INSTALL_DIR>\deployment_tools\open_model_zoo\demos\human_pose_estimation_3d_demo\python

Then, you should be able to run the demo by executing the following command:

python3 human_pose_estimation_3d_demo.py -m <path_to_model>/human-pose-estimation-3d-0001.xml -i <path_to_video>/video_name.mp4

Regards,

Wan

[thunfischtoast]

Hi Wan,

thank you again for providing help!

After copying the files the demo works.

I guess we should update the manual / readme as this info is missing? Or I missed it.

Cheers,

Christian

[Wan_Intel]

Hi Christian,

Thanks for your suggestion!

This thread will no longer be monitored since this issue has been resolved. 

If you need any additional information from Intel, please submit a new question.

Cheers,

Wan