Python: bodypix model from tfjs: cannot create layer Minimum in getLayerInstance

[diddledani]

System information (version)

• OpenCV => opencv_python-4.2.0.32 (inside a venv installed via pip) and also a version built from master (commit hash 0979940)
• Operating System / Platform => Ubuntu 20.04 (amd64)
• Compiler => ❔ N/A

Detailed description

The converted bodypix (mobilenet v2?) dnn model from tfjs loaded into opencv in python using cvNet = cv2.dnn.readNetFromTensorflow fails when calling cvNet.forward() with the following error:

Traceback (most recent call last):
  File "test-bodypix.py", line 7, in <module>
    results = cvNet.forward()
cv2.error: OpenCV(4.2.0) /io/opencv/modules/dnn/src/dnn.cpp:562: error: (-2:Unspecified error) Can't create layer "clip_by_value/Minimum" of type "Minimum" in function 'getLayerInstance'

Download URLs for original unconverted bodypix model.json files (I've also attached the converted model.pb that I am using below):

https://storage.googleapis.com/tfjs-models/savedmodel/bodypix/mobilenet/float/050/model-stride16.json
https://storage.googleapis.com/tfjs-models/savedmodel/bodypix/mobilenet/float/050/group1-shard1of1.bin

Steps to reproduce

• model.pb and minimum python implementation for reproduction: bodypix-test.zip
• just the model.pb: model.zip

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues,
  answers.opencv.org, Stack Overflow, etc and have not found solution
• I updated to latest OpenCV version and the issue is still there
• There is reproducer code and related data files: videos, images, onnx, etc

[dkurt]

opencv_python-4.2.0.32

Please try the latest 4.3.0

[diddledani]

I edited my post before you replied to indicate that I also tried master branch. Also note that 4.3.0 does not exist in the pypi.org package repository.

[dkurt]

@diddledan, oh, sorry, just missed it. Thanks!

[diddledani]

No problem - I figured I edited it after you'd loaded the page so it was unobvious :-)

[jtg-gg]

I am having the same issue, is there anyway to fix this ?

[jtg-gg]

I manage to draw the graph using tensor board. The Minimum / Maximum operation is actually at the very last operation for "float_parts_offsets"
@dkurt has implemented some graph transformation for fusing this into Relu6 ?
I am not sure why it isn't for this case

[dkurt]

@l-bat, please take a look

[dkurt]

Please test #17386 by @l-bat