Encountered unknown input type of a loop v5::Loop

[warce]

OS: Windows 10

Inference Engine 2021.3

Model: TensorFlow 2 custom model

I'm trying to convert a custom TensorFlow 2 model from an IR file into a Myriad blob.  The Model Optimizer successfully compiles the model into an IR version 10 representation with this command: 

"C:\Program Files (x86)\Intel\openvino_2021.3.394\deployment_tools\model_optimizer\mo_tf.py" --saved_model_dir "{path to model}" --input_shape=[1,30,160,160,3] --data_type=FP16 --mean_values=[127.5,127.5,127.5] --scale_values=[255,255,255] --static_shape --log_level=DEBUG

When I attempt to convert it from the IR file into a Myriad blob, I get the following error:

Inference Engine:
        API version ............ 2.1
        Build .................. 2021.3.0-2787-60059f2c755-releases/2021/3
        Description ....... API

Network inputs:
    time_distributed_input : FP16 / NCDHW
Network outputs:
    StatefulPartitionedCall/sequential_1/dense_2/Sigmoid : FP16 / NC

[Warning][VPU][Config] Deprecated option was used : VPU_MYRIAD_PLATFORM
Encountered unknown input type of a loop v5::Loop StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/while (Copy_StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/strided_slice/Output_0/Data__const[0]:i32{}, StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/while/ExecutionConditionValue/Output_0/Data__const[0]:boolean{}, StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/time/Output_0/Data__const[0]:i32{}, Func/StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/input/_45/Output_0/Data__const[0]:f16{1,30}, Func/StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/input/_46/Output_0/Data__const[0]:f16{1,30}, StatefulPartitionedCall/sequential_1/lstm/PartitionedCall/transpose[0]:f16{30,1,12800}) -> (f16{30,1,30}) at index 0

Is the loop input type not supported?  The model uses the Keras TimeDistributed as two layers, which is where the loop should be coming from. 

[Peh_Intel · Accepted solution]

Hi warce,

Thanks for reaching out to us.

For your information, this Loop-5 operation is newly added support by CPU plugin starting from OpenVINO™ Toolkit 2021.2. However, it is still not yet supported by MYRIAD plugin. This information is available in the Release Note, under the sub-section “Inference Engine” of the section “New and Changed in the Release 2”.

Regards,

Peh

[Peh_Intel]

Hi warce,

This thread will no longer be monitored since we have provided the answer. If you need any additional information from Intel, please submit a new question. 

Regards,

Peh