I am using OpenVINO 2021.2 on Windows 10.
I am facing the below issue during inference:
Interpolate operation should be converted to Interp
My original model is TensorFlow PB, I can convert the model using model optimizer without any problem.
Model inference will run successfully if NHWC to NCHW is enabled. but if I disabled this channel conversion, Inference will fail.
python mo_tf.py --input_model ./model.pb --input input_3 --output conv2d_72/Sigmoid --input_shape [1,256,256,1]
Infernce Fail Command:
python mo_tf.py --input_model ./model.pb --input input_3 --output conv2d_72/Sigmoid --input_shape [1,256,256,1] --disable_nhwc_to_nchw
"Both conversion commands, Model Optimizer runs correctly and produce IR". but when I add `--disable_nhwc_to_nchw` (command 2) I will not be able to infer correctly and I will get following error: `Interpolate operation should be converted to Interp `.
The reason for this is that Tensorflow is the only network out there that uses NHWC. All other networks use NCHW. Thus OpenVino's Inference Engine satisfies the majority so it also uses NCHW layout. Models of all network flavors must be converted to NCHW layout in order to work with Inference Engine.
Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question.