When using Keras' ResNet50 model with the top (softmax) layer removed and adding a Dense layer, the model optimiser is able to convert the resultant Tensorflow model.
Output from Keras model and Tensorflow model are identical as expected.
Output from Inference Engine is quite different.
I have created a code sample to reproduce the issue. Please see attached sample.
Python dependencies are listed in requirements.txt
To generate Tensorflow and Keras dummy models:
This will generate dummy models:
Generate MO model from models/tf/dummy.pb and place them in models/mo/dummy/FP32:
/opt/intel/computer_vision_sdk/deployment_tools/model_optimizer$ sudo python3 mo.py --input_model PATH_TO_EXTRACTED_FOLDER/models/tf/dummy.pb --input_shape "(1,182,182,3)"
Now place the generated models in models/mo/dummy/FP32.
Run the test script. You can see that the output is different.