Running single pose estimation demo

[jain__Yasha]

I am trying to run the single pose estimation demo model using 

single_human_pose_estimation_demo.py --model_od mobilenet-ssd.xml --model_hpe human-pose-estimation-0001.xml

I keep getting error 

\Desktop\single_human_pose_estimation_demo\estimator.py", line 92, in __init__
    assert len(self.model.outputs) == 1, "Expected 1 output blob"
AssertionError: Expected 1 output blob

Why is this happeing

[SIRIGIRI_V_Intel]

Hi Yasha,

To run the single-human-pose-estimation-demo, use single-human-pose-estimation-0001 model which is available to download using model downloader.

you can follow the below commands to convert the single-human-pose-estimation-0001.pth to IR files.:

python3 /opt/intel/openvino/deployment_tools/tools/model_downloader/pytorch_to_onnx.py --model-name=SinglePersonPoseEstimationWithMobileNet --weights=single-human-pose-estimation-0001.pth --input-shape=1,3,384,288 --output-file=single-human-pose-estimation-0001.onnx --output-names=heatmaps0,heatmaps1,heatmaps2,heatmaps3,heatmaps4,heatmaps --input-names=data --import-module=models

python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo_onnx.py --input=data --mean_values=data[123.675,116.28,103.53] --scale_values=data[58.395,57.12,57.375] --output=heatmaps --input_model=single-human-pose-estimation-0001.onnx

Then the run the single-human-pose-estimation-demo using the generated IR and object-detection model.

Hope this helps.

Regards,

Ram prasad

[jain__Yasha]

Hi , running the first script gives an error : pytorch_to_onnx.py: error: unrecognized arguments: single-human-pose-estimation-0001.onnx ?

[SIRIGIRI_V_Intel]

It seems there is a space in the first command mentioned. I have updated the command.

Can you please try once again?

Regards,

Ram prasad

[jain__Yasha]

Hi Ram, 

Thank you so much for your help! it works

Sadly the detected FPS with NCS2 is less than 1.6 FPS so almost not usable. 

Are there any models that work at higher FPS?

[SIRIGIRI_V_Intel]

Hi Yasha,

This is the only model that works for this demo.

The OpenVINO models are for the demonstration purpose. You can train and use your models for more accuracy.

Feel free to ask any other questions.

Regards,

Ram prasad

[wang__huiyong]

python3 /opt/intel/openvino/deployment_tools/tools/model_downloader/pytorch_to_onnx.py --model-name=SinglePersonPoseEstimationWithMobileNet --weights=single-human-pose-estimation-0001.pth --input-shape=1,3,384,288 --output-file=single-human-pose-estimation-0001.onnx --output-names=heatmaps0,heatmaps1,heatmaps2,heatmaps3,heatmaps4,heatmaps --input-names=data --import-module=models

Module SinglePersonPoseEstimationWithMobileNet doesn't exist. Check if it is installed
No module named 'models'   

?????

[SIRIGIRI_V_Intel]

Can you try to export the path to models directory using PYTHONPATH as mentioned below:

export PYTHONPATH=/opt/intel/openvino/deployment_tools/tools/model_downloader/public/single-human-pose-estimation-0001/single-human-pose-estimation-0001

Hope this helps.

Regards,

Ram prasad

[DylanPoh]

Module SinglePersonPoseEstimationWithMobileNet doesn't exist. Check if it is installed
No module named 'models'   

---

@SIRIGIRI_V_Intel wrote:

Can you try to export the path to models directory using PYTHONPATH as mentioned below:

export PYTHONPATH=/opt/intel/openvino/deployment_tools/tools/model_downloader/public/single-human-pose-estimation-0001/single-human-pose-estimation-0001

Hope this helps.

Regards,

Ram prasad

---

Hi, I have similar errors, but I am running in Windows 10 environment. How can I solve this since export or set pythonpath is not working for me

[DylanPoh]

Hi, I have similar errors, but I am running in Windows 10 environment. How can I solve this since export or set pythonpath is not working for me

[veerbhansari]

Hello,

Were you able to figure it out? Even I am using it in windows 10 environment and getting the same error.

Thanks!

[zhang__zihao]

Ram prasad (Intel) wrote:

Hi Yasha,

To run the single-human-pose-estimation-demo, use single-human-pose-estimation-0001 model which is available to download using model downloader.

you can follow the below commands to convert the single-human-pose-estimation-0001.pth to IR files.:

python3 /opt/intel/openvino/deployment_tools/tools/model_downloader/pytorch_to_onnx.py --model-name=SinglePersonPoseEstimationWithMobileNet --weights=single-human-pose-estimation-0001.pth --input-shape=1,3,384,288 --output-file=single-human-pose-estimation-0001.onnx --output-names=heatmaps0,heatmaps1,heatmaps2,heatmaps3,heatmaps4,heatmaps --input-names=data --import-module=models

python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo_onnx.py --input=data --mean_values=data[123.675,116.28,103.53] --scale_values=data[58.395,57.12,57.375] --output=heatmaps --input_model=single-human-pose-estimation-0001.onnx

Then the run the single-human-pose-estimation-demo using the generated IR and object-detection model.

Hope this helps.

Regards,

Ram prasad

Hello Ram

When I run 

python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo_onnx.py --input=data --mean_values=data[123.675,116.28,103.53] --scale_values=data[58.395,57.12,57.375] --output=heatmaps --input_model=single-human-pose-estimation-0001.onnx

It shows 

zsh: no matches found: --mean_values=data[123.675,116.28,103.53]

What should I do?

(MacOS, Openvino 2020.3)

[Max_L_Intel]

zhang, zihao wrote:

Hello Ram

When I run 

python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo_onnx.py --input=data --mean_values=data[123.675,116.28,103.53] --scale_values=data[58.395,57.12,57.375] --output=heatmaps --input_model=single-human-pose-estimation-0001.onnx

It shows 

zsh: no matches found: --mean_values=data[123.675,116.28,103.53]

What should I do?

(MacOS, Openvino 2020.3)

Try to run the complete command

python3 --input=data --mean_values=data[123.675,116.28,103.53] --scale_values=data[58.395,57.12,57.375] --output=heatmaps --input_model=single-human-pose-estimation-0001.onnx --reverse_input_channels --input_shape=[1,3,384,288]

Or if you are using the original PyTorch model of single-human-pose-estimation-0001, you could use the converter.py within openvino/deployment_tools/open_model_zoo/tools/downloader/ for full conversion over to ONNX and then further to IR.

Hope this helps.
Thanks.