[Recurrent layers of dnn module] fully support to LSTM and GRU is needed

[HannibalAPE]

System information (version)

• OpenCV => 4.5
• Operating System / Platform => Ubuntu 20.04
• Compiler => ❔

Detailed description

1. Currently, the LSTM of ONNX importer can not support the non-zero inputs of hidden state and cell state.
   

   opencv/modules/dnn/src/onnx/onnx_importer.cpp

   Line 904 in 941a979

   CV_CheckEQ(countNonZero(getBlob(node_proto, 5)), 0, "Unsupported non zero initial_h");

   
   

   opencv/modules/dnn/src/onnx/onnx_importer.cpp

   Line 905 in 941a979

   CV_CheckEQ(countNonZero(getBlob(node_proto, 6)), 0, "Unsupported non zero initial_c");

2. ONNX importer can not support GRU layer.
   When can this layer be added to the ONNX importer?
   It also needs to support the non-zero inputs of the hidden state.

These features are needed in the state-of-the-art methods of Text Recognition.
Thank you in advance!

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

[JulieBar]

@HannibalAPE it should work now. Please try latest build

[asmorkalov]

@HannibalAPE Friendly reminder.

[HannibalAPE]

@JulieBar @asmorkalov Thanks a lot. We will use them to build some modern text recognition networks.

[asmorkalov]

#23475

[asmorkalov]

#23501