Bilinear upsampling layer shift

[aledenza]

System information

• OS Platform and Distribution: Windows 10 x64
• TensorFlow installed from: pypy
• TensorFlow version: 2.3.0
• Python version: 3.8.5

Describe the current behavior
Inference with Opencv Dnn has Mask shift that apperas while tensorflow model inference, if model has Resize Bilinear Layer. There's no shift if ResizeBilinear layer replaced with ResizeNearest layer. Also tensorflow inference engine does not have this issue and works well with Bilinear and Nearest resize.
I agree, that this looks like Opencv issue, but i found this bug (#29856) that took place in TF 2.0 beta and it looks similar to this issue. Now tensorflow inference works fine, but, probably, only inference was fixed, but layer still works unproperly.

Here you can see results of inference:

Inference results

OpenCV DNN Bilinear

OpenCV DNN Nearest

Tensorflow Bilinear

Tensorflow Nearest

Steps to reproduce

reproduce.zip

[Saduf2019]

@ZakharovDenis
I ran the code shared and face a different issue, please find the gist here.

[aledenza]

Hello @Saduf2019 . You have to upload model to colab storage to run the code. Download link provided in download_models.txt file. I can't put models to the zip file due to github file size limit.
I updated a reproduce.zip to run opencv segmentation from python, not from c++.

[Saduf2019]

@ZakharovDenis
Apologies for the delayed response, i ran the code and face this error

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler]

Closing as stale. Please reopen if you'd like to work on this further.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No