I am using pixelshuffle in my tensorflow neural network model and I used Openvino converting the model to .xml and .bin files.
All works fine.
However, the pixelshuffle does not seem work properly. Instead channel data fill into pixel data, the channels are concating each other.
e.g. 4 channels with pixelshuffle becomes
while each pixel should be filled (expanded) with pixels from 4 channels instead.
I am using Intel Neural Compute Stick v1 and Openvino python sdk.
Any one has similar problem?
For more complete information about compiler optimizations, see our Optimization Notice.