an error when running an optimizer

[kaishijeng]

Got an error when running optimizer below. Any idea why this happens:

python /opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/mo.py --input_model human-pose-estimation-3d.onnx --input=data --mean_values=data[128.0,128.0,128.0] --scale_values=data[255.0,255.0,255.0] --output=features,heatmaps,pafs

Model Optimizer arguments:
Common parameters:
- Path to the Input Model: /home/user/2TB/openvino/lightweight-human-pose-estimation-3d-demo.pytorch/human-pose-estimation-3d.onnx
- Path for generated IR: /home/user/2TB/openvino/lightweight-human-pose-estimation-3d-demo.pytorch/.
- IR output name: human-pose-estimation-3d
- Log level: ERROR
- Batch: Not specified, inherited from the model
- Input layers: data
- Output layers: features,heatmaps,pafs
- Input shapes: Not specified, inherited from the model
- Mean values: data[128.0,128.0,128.0]
- Scale values: data[255.0,255.0,255.0]
- Scale factor: Not specified
- Precision of IR: FP32
- Enable fusing: True
- Enable grouped convolutions fusing: True
- Move mean values to preprocess section: False
- Reverse input channels: False
ONNX specific parameters:
Model Optimizer version: 2020.2.0-60-g0bc66e26ff
[ ERROR ] Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.user_data_repack.UserDataRepack'>): No node with name features.
For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #51.

[Daniil-Osokin]

Hi, No node with name features means that MO did not found features node, which was specified as one of output node above. You can check if onnx model has it (by looking at conversion output). Did you use scripts/convert_to_onnx.py for conversion to onnx?

[kaishijeng]

Yes, I used scripts/convert_to_onnx.py for conversion to onnx.
I grep features from conversion output and it seems features is there:

%628 : Float(1, 128, 32, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%627, %Pose3D.prediction.feature_maps.0.0.weight, %Pose3D.prediction.feature_maps.0.0.bias) # /home/fc/anaconda3/envs/directron2/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
%features : Float(1, 57, 32, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%629, %Pose3D.prediction.feature_maps.1.0.weight, %Pose3D.prediction.feature_maps.1.0.bias) # /home/fc/anaconda3/envs/directron2/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
return (%features, %heatmaps, %pafs)

[Daniil-Osokin]

So it looks like an internal error in MO. Try to use official model downloader and converter, model name is human-pose-estimation-3d-0001.

[kaishijeng]

I will download the model and try it out.

Thanks

[ZlodeiBaal]

Same problem here.
Fix it with roll back to torch 1.4 version (from 1.5).
"human-pose-estimation-3d-0001" and OpenVino affected as well.

[Daniil-Osokin]

Thanks for the solution, @ZlodeiBaal!

[Daniil-Osokin]

@ZlodeiBaal BTW, Anton, really love your posts on habr! 👑

[ZlodeiBaal]

@Daniil-Osokin
Thank you, Daniil! You have a really cool git! Already several times while googling I have found answers and interesting ideas here :)