Hi i am trying to re-train the Keras inception v3 model then converting to Tensorflow and later compiling to intel NCS. The Tensorflow model is tested with the acompanying weight file and gives the expected output probabilities. However the NCS forward pass does not. Looking back at the network diagram from the NCS profile tool i noticed that batch norm layers and conv layers are being omitted from the compilation. Other networks like VGG16 without batch norm work fine. If this issue is resolved, in my opinion it will add great value to the comunity regarding Keras support.
