convert to onnx https://github.com/MaybeShewill-CV/CRNN_Tensorflow

[AR92]

hi,

i am following this link https://docs.openvinotoolkit.org/2020.1/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_CRNN_From_Tensorflow.html 

in the third step i was getting AssertionError  

frozen = tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['shadow/LSTMLayers/transpose_time_major'])

 shadow/LSTMLayers/transpose_time_major is not in graph

i changed it to below 

frozen = tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['shadow_net/sequence_rnn_module/stack_bidirectional_rnn/cell_0/bidirectional_rnn/fw/fw/while/Identity_2'])

and frozen_graph is generated, i am trying to use this frozen graph with this link example https://github.com/opencv/opencv/blob/master/samples/dnn/text_detection.cpp 

and opencv throws error

[ERROR:0] global C:\jenkins\workspace\OpenCV\OpenVINO\2020.3\build\windows\opencv\modules\dnn\src\dnn.cpp (3272) cv::dnn::dnn4_v20200310::Net::Impl::getLayerShapesRecursively OPENCV/DNN: []:(_input): getMemoryShapes() throws exception. inputs=1 outputs=0/0 blobs=0
[ERROR:0] global C:\jenkins\workspace\OpenCV\OpenVINO\2020.3\build\windows\opencv\modules\dnn\src\dnn.cpp (3275) cv::dnn::dnn4_v20200310::Net::Impl::getLayerShapesRecursively input[0] = [ 1 1 32 100 ]
[ERROR:0] global C:\jenkins\workspace\OpenCV\OpenVINO\2020.3\build\windows\opencv\modules\dnn\src\dnn.cpp (3285) cv::dnn::dnn4_v20200310::Net::Impl::getLayerShapesRecursively Exception message: OpenCV(4.3.0-openvino-2020.3.0) C:\jenkins\workspace\OpenCV\OpenVINO\2020.3\build\windows\opencv\modules\dnn\src\dnn.cpp:790: error: (-215:Assertion failed) inputs.size() == requiredOutputs in function 'cv::dnn::dnn4_v20200310::DataLayer::getMemoryShapes'

OpenCV: terminate handler is called! The last OpenCV error is:
OpenCV(4.3.0-openvino-2020.3.0) Error: Assertion failed (inputs.size() == requiredOutputs) in cv::dnn::dnn4_v20200310::DataLayer::getMemoryShapes, file C:\jenkins\workspace\OpenCV\OpenVINO\2020.3\build\windows\opencv\modules\dnn\src\dnn.cpp, line 790

i have trained my model from https://github.com/MaybeShewill-CV/CRNN_Tensorflow.git. 

i am new to tensorflow , can anyone please have a look  

i have uploaded my model to drive https://drive.google.com/drive/folders/1wgFcC3a5jMqcRFvKmj4XFAv_b7ATt9xV?usp=sharing please have a look.

[AR92]

i also tried to convert it into IR, but got some error

Model Optimizer version:
2020-07-28 17:18:42.925656: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_100.dll'; dlerror: cudart64_100.dll not found
2020-07-28 17:18:42.925785: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.output_cut.OutputCut'>): Graph contains 0 node after executing <class 'extensions.front.output_cut.OutputCut'>. It considered as error because resulting IR will be empty which is not usual

[Munesh_Intel]

Hi Amit,

Please provide the command given to Model Optimizer to convert your trained model to Intermediate Representation (IR), and environment details (versions of OS, TensorFlow, Python, CMake, etc.).

Regards,

Munesh

[AR92]

hi ,

command i tried is 

python mo_tf.py --input_model E:\CRNN_Tensorflow-case\frozen_graph_test.pb --reverse_input_channels   --data_type FP32 --input_shape="[1,32,100,3]"  --output_dir IR

 i trained this model in ubuntu 18.04, tensoflow-gpu 1.15, python 3.7

and to convert the frozen_graph i used

windows 10 x64bit machine,

tensorflow-cpu 1.15,

python 3.7

cmake 3.14.5

i have also converted a tensorflow ssd object detection model to IR and that is working fine.

[Sahira_Intel · Accepted solution]

Hi,

https://github.com/MaybeShewill-CV/CRNN_Tensorflow has not been validated for the C++ Text Detection sample.

Best Regards,

Sahira